<%BANNER%>

Impact of Electrically and Thermally Induced Physical Defects on the Reliability of AlGaN/GaN High Electron Mobility Tra...

MISSING IMAGE

Material Information

Title:
Impact of Electrically and Thermally Induced Physical Defects on the Reliability of AlGaN/GaN High Electron Mobility Transistors
Physical Description:
1 online resource (192 p.)
Language:
english
Creator:
Holzworth, Monta Raymond, Jr
Publisher:
University of Florida
Place of Publication:
Gainesville, Fla.
Publication Date:

Thesis/Dissertation Information

Degree:
Doctorate ( Ph.D.)
Degree Grantor:
University of Florida
Degree Disciplines:
Materials Science and Engineering
Committee Chair:
Jones, Kevin S
Committee Members:
Norton, David P
Law, Mark E
Pearton, Stephen J
Ren, Fan

Subjects

Subjects / Keywords:
algan -- gan -- hemt -- reliability
Materials Science and Engineering -- Dissertations, Academic -- UF
Genre:
Materials Science and Engineering thesis, Ph.D.
bibliography   ( marcgt )
theses   ( marcgt )
government publication (state, provincial, terriorial, dependent)   ( marcgt )
born-digital   ( sobekcm )
Electronic Thesis or Dissertation

Notes

Abstract:
AlGaN/GaN high electron mobility transistors are unique for their combination of high temperature, high power, and high frequency applications.  Compared to Si, Ge, and compound semiconductors such as GaAS and InP, AlGaN/GaN transistors outclass the current technology due to their superior combination of high breakdown voltage and high frequency performance.  These characteristics arise from structural and electrical properties inherent to the AlGaN/GaN heterojunction which have enabled AlGaN/GaN transistors usage in important military and civilian applications such as microwave and millimeter technology, RADAR systems, and as high current and voltage switches in utility grid systems.  As the technology continues to improve due to increased materials quality and device advancements, future applications will require AlGaN/GaN transistor usage under even higher voltages and temperatures.  Therefore, the effects of these stresses need to be investigated in order improve device performance and reliability. When stress conditions are applied in combination, device failure is accelerated.  However, future applications may require electrical or thermal stress to be effectively applied separately.  Therefore, it is important to understand how each factor contributes individually to the reliability and failure mechanisms in transistors to determine the actual working device life-time in its operating environment.  The following research employs structural, chemical, and electrical device characterization paired with simulation in order to develop structure-property relationships between defects and device performance.  Here, for the first time, as-grown gate interfacial layers were characterized using atom probe tomography and are composed of two distinct oxide layers, NiOx and AlOx.  Knowing the composition permits elimination or reduction of the layers through etching or processing modifications.  Furthermore, using off-state reverse bias electrical stress, Ni-gate metal reactions with AlGaN epilayers emulate the shape and size of the electric field contours between 5 – 6 MV/cm; an advancement from the previous understanding that defects form only at the peak electric field.  Finally, application of only thermal stress is shown for the first time to cause gate metal penetration into threading dislocations.  This penetration combined with Ni-Au gate metal interdiffusion causes a negative shift in threshold voltage, an increase in drain current, and only above 400°C an increase in gate leakage.
General Note:
In the series University of Florida Digital Collections.
General Note:
Includes vita.
Bibliography:
Includes bibliographical references.
Source of Description:
Description based on online resource; title from PDF title page.
Source of Description:
This bibliographic record is available under the Creative Commons CC0 public domain dedication. The University of Florida Libraries, as creator of this bibliographic record, has waived all rights to it worldwide under copyright law, including all related and neighboring rights, to the extent allowed by law.
Statement of Responsibility:
by Monta Raymond Holzworth.
Thesis:
Thesis (Ph.D.)--University of Florida, 2013.
Local:
Adviser: Jones, Kevin S.

Record Information

Source Institution:
UFRGP
Rights Management:
Applicable rights reserved.
Classification:
lcc - LD1780 2013
System ID:
UFE0045931:00001

MISSING IMAGE

Material Information

Title:
Impact of Electrically and Thermally Induced Physical Defects on the Reliability of AlGaN/GaN High Electron Mobility Transistors
Physical Description:
1 online resource (192 p.)
Language:
english
Creator:
Holzworth, Monta Raymond, Jr
Publisher:
University of Florida
Place of Publication:
Gainesville, Fla.
Publication Date:

Thesis/Dissertation Information

Degree:
Doctorate ( Ph.D.)
Degree Grantor:
University of Florida
Degree Disciplines:
Materials Science and Engineering
Committee Chair:
Jones, Kevin S
Committee Members:
Norton, David P
Law, Mark E
Pearton, Stephen J
Ren, Fan

Subjects

Subjects / Keywords:
algan -- gan -- hemt -- reliability
Materials Science and Engineering -- Dissertations, Academic -- UF
Genre:
Materials Science and Engineering thesis, Ph.D.
bibliography   ( marcgt )
theses   ( marcgt )
government publication (state, provincial, terriorial, dependent)   ( marcgt )
born-digital   ( sobekcm )
Electronic Thesis or Dissertation

Notes

Abstract:
AlGaN/GaN high electron mobility transistors are unique for their combination of high temperature, high power, and high frequency applications.  Compared to Si, Ge, and compound semiconductors such as GaAS and InP, AlGaN/GaN transistors outclass the current technology due to their superior combination of high breakdown voltage and high frequency performance.  These characteristics arise from structural and electrical properties inherent to the AlGaN/GaN heterojunction which have enabled AlGaN/GaN transistors usage in important military and civilian applications such as microwave and millimeter technology, RADAR systems, and as high current and voltage switches in utility grid systems.  As the technology continues to improve due to increased materials quality and device advancements, future applications will require AlGaN/GaN transistor usage under even higher voltages and temperatures.  Therefore, the effects of these stresses need to be investigated in order improve device performance and reliability. When stress conditions are applied in combination, device failure is accelerated.  However, future applications may require electrical or thermal stress to be effectively applied separately.  Therefore, it is important to understand how each factor contributes individually to the reliability and failure mechanisms in transistors to determine the actual working device life-time in its operating environment.  The following research employs structural, chemical, and electrical device characterization paired with simulation in order to develop structure-property relationships between defects and device performance.  Here, for the first time, as-grown gate interfacial layers were characterized using atom probe tomography and are composed of two distinct oxide layers, NiOx and AlOx.  Knowing the composition permits elimination or reduction of the layers through etching or processing modifications.  Furthermore, using off-state reverse bias electrical stress, Ni-gate metal reactions with AlGaN epilayers emulate the shape and size of the electric field contours between 5 – 6 MV/cm; an advancement from the previous understanding that defects form only at the peak electric field.  Finally, application of only thermal stress is shown for the first time to cause gate metal penetration into threading dislocations.  This penetration combined with Ni-Au gate metal interdiffusion causes a negative shift in threshold voltage, an increase in drain current, and only above 400°C an increase in gate leakage.
General Note:
In the series University of Florida Digital Collections.
General Note:
Includes vita.
Bibliography:
Includes bibliographical references.
Source of Description:
Description based on online resource; title from PDF title page.
Source of Description:
This bibliographic record is available under the Creative Commons CC0 public domain dedication. The University of Florida Libraries, as creator of this bibliographic record, has waived all rights to it worldwide under copyright law, including all related and neighboring rights, to the extent allowed by law.
Statement of Responsibility:
by Monta Raymond Holzworth.
Thesis:
Thesis (Ph.D.)--University of Florida, 2013.
Local:
Adviser: Jones, Kevin S.

Record Information

Source Institution:
UFRGP
Rights Management:
Applicable rights reserved.
Classification:
lcc - LD1780 2013
System ID:
UFE0045931:00001


This item has the following downloads:


Full Text

PAGE 1

1 IMPACT OF ELECTRICALLY AND THERMALLY INDUCED PHYSICAL DEFECTS ON THE RELIABILITY OF ALGAN/GAN HIGH ELECTRON MOBILITY TRANSISTORS By MONTA RAYMOND HOLZWORTH, JR. A DISSERTATION PRESENTED TO THE GRADUATE SCHOOL OF THE UNIV ERSITY OF FLORIDA IN PARTIAL FULFILLMENT OF THE REQUIREMENTS FOR THE DEGREE OF DOCTOR OF PHILOSOPHY UNIVERSITY OF FLORIDA 2013

PAGE 2

2 2013 Monta Raymond Holzworth, Jr.

PAGE 3

3 To Jennette, M.R. III, and Tyson Sabertoo th Tigerbear Holzworth

PAGE 4

4 ACKNOWLEDGMENTS I would like t o acknowledge my family for all the love and support they have given me over the years. Without them this endeavor would have been a much more en possible. Specifically, I would like to thank my mom and dad for allowing me to pursue any academic interest that I wanted to follow and for always being encouraging While it may take a village to raise a child, it takes exposure to many different di sciplines, people, cultures, and environments to become open minded enough to question foundations of thoughts, beliefs, values, and theories I would like to acknowledge the specific sacrifices my wife has made for me /us while I pur sue d this degree. These range from dealing with the odd and long hour work schedule, budgeting and counting every last penny, and being separated by an ocean as I interned in Japan. She has always been supportive of me and my work and I am forever thank ful and grateful I I would like to thank my advisor Prof. Kevin Jones for offering me an original and interesting research project when I joined his gr oup and for giving me the freedom to explore the subject in any direction His advice and suggestions always inspired questions into a more fundamental and complete understanding of the problems being investigated. I would like to thank my committee mem bers Prof. Stephen Pearton, Prof. Fan Ren, Prof. Mark Law, and Prof. David Norton for always taking the time to answer my questions, offer ing guidance and collabor ative research, and for the time investment of being on this committee.

PAGE 5

5 Additionally, I ack nowledge my former and current group members whom I have seen on a daily basis and have provided me friendship and lab assistance: Lucia Romano, Nick Rudawski, Dan Gostovic, Leah Edelman, Sam Moore, Michell e Phen, Sidan Jin, Blake Darby, Nick Vito, Brad Ya tes, Patrick Whiting, Aaron Lind Tom Martin, and Henry Aldridge I would also like to thank my fellow collaborators for their assistance: Liu Lu, Tsung Sheng Kang, Erica Douglas, Dave Cheney, Erin Patrick, David Horton, and Weikai Xu. Finally I would li ke to acknowledge Dr. Kazuhiro Hono san of the Magnetic Materials Unit at the National Institute for Materials Science in Tsukuba, Japan. I will always be grateful for the four month internship, 80+ hours of machine time per week up to try to get atom probe tomography to work, and his time whenever I would stop by his office. Additionally, I thank Dr. Ohkubo san, Dr. Kodzuka san, Dr. Sasaki san, Dr. Mendis, and particularly Dr. Navid Sep ehri, for the much appreciated technical assist ance and friendship while I was in Japan. I will always fondly remember my time spent there.

PAGE 6

6 TABLE OF CONTENTS page ACKNOWLEDGMENTS ................................ ................................ ................................ .. 4 LIST OF TABLES ................................ ................................ ................................ ............ 9 LIST OF FIGURES ................................ ................................ ................................ ........ 10 LIST OF ABBREVIATIONS ................................ ................................ ........................... 14 ABSTRACT ................................ ................................ ................................ ................... 16 CHAPTER 1 INTRODUCTION ................................ ................................ ................................ .... 18 Motivation ................................ ................................ ................................ ............... 18 Objectives of Research ................................ ................................ ........................... 19 Dissertation Outline ................................ ................................ ................................ 19 2 BACKGROUND ................................ ................................ ................................ ...... 21 2.1 Properties of GaN and AlGaN ................................ ................................ .......... 21 2.2 Transistor Fa brication ................................ ................................ ...................... 24 2.3 Device Physics ................................ ................................ ................................ 28 2.3.1 Schottky Contacts ................................ ................................ .................... 28 2 .3.2 Ohmic Contacts ................................ ................................ ....................... 29 2.3.3 Device Operation ................................ ................................ ..................... 29 2.4 Reliability and Degradation Issues in AlGaN/GaN HEMTs .............................. 30 2.4.1 Surface Traps ................................ ................................ .......................... 31 2.4.2 Buffer Traps ................................ ................................ ............................. 31 2.4.3 Hot Electron Effects ................................ ................................ ................. 32 2.4.4 Gate Contact Degradation ................................ ................................ ....... 33 2.4.5 Source and Drain Contact Degradation ................................ ................... 34 2.4.6 Inverse Piezoelectric Effect ................................ ................................ .... 35 2.4.7 Threading Dislocation Defects and Impurity Diffusion in AlGaN/GaN HEMTs ................................ ................................ ................................ .......... 38 2.4.8 Physical Degradation in Pt Gate AlGaN/GaN/HEMTs ............................ 40 2.4.9 Physical Degradation in Ni Gate AlGaN/GaN/HEMTs ............................ 47 3 EXPERIM ENTAL TECHNIQUES ................................ ................................ ............ 51 3.1 Material Processing ................................ ................................ ......................... 51 3.1.1 Thermal Annealing ................................ ................................ .................. 51 3.1.2 Transistor Etching ................................ ................................ ................... 52 3.1.3 XTEM Sample Preparation ................................ ................................ ...... 53

PAGE 7

7 3.1.4 APT Sample Preparation ................................ ................................ ......... 55 3.1.4.1 Si post array mounting technique ................................ ................... 55 3.1.4.2 W wire mounting technique ................................ ............................ 56 3.2 Ma terial Characterization ................................ ................................ ................. 56 3.2.1 Electrical ................................ ................................ ................................ .. 56 3.2.2 SEM ................................ ................................ ................................ ......... 57 3.2. 3 TEM ................................ ................................ ................................ ......... 58 3.2.4 APT ................................ ................................ ................................ ......... 61 4 CHARACTERIZATION OF GATE INTERFACIAL LAYERS ................................ ... 70 4.1 Effect of Interfacial Layers on HEMTs ................................ ............................... 70 4.2 Experimental ................................ ................................ ................................ ..... 71 4.2.1 Device Structures ................................ ................................ .................... 71 4.2.1.1 Si(111) substrate Ni gate HEMT ................................ .................... 71 4.2.1.2 SiC substrate Ni gate HEMT ................................ .......................... 71 4.2.1.3 Pt ga te HEMT ................................ ................................ ................ 72 4.2.2 APT Systems ................................ ................................ ........................... 72 4.2.2.1 Image LEAP tomography system ................................ ................... 72 4.2.2.1 Cameca APT system ................................ ................................ ..... 72 4.3 Results and Discussion ................................ ................................ ..................... 73 4.3.1 Structural TEM Analysis ................................ ................................ .......... 73 4.3.2 Imago APT Analysis ................................ ................................ ................ 73 4.3.3 Cameca APT Analysis ................................ ................................ ............. 77 4.3.4 Aberration Corrected XTEM Che mical Analysis ................................ ...... 81 4.4 Summary ................................ ................................ ................................ .......... 82 5 EFFECT OF ELECTRIC FIELDS ON RELIABILILITY ................................ ............ 92 5.1 Need to Understand Physical Defects in AlGaN/GaN HEMTs .......................... 92 5.2 Experimental ................................ ................................ ................................ ..... 93 5.2.1 Device Structure ................................ ................................ ...................... 93 5.2.2 Electrical Stress Conditions ................................ ................................ ..... 93 5.3 Results and Discussion ................................ ................................ ..................... 94 5.3.1 Relationship between I Dmax and Large Scale Physical Degradation ........ 94 5.3.2 Relationship between I G and Large Scale Physical Degradation ........... 106 5.4 Summary ................................ ................................ ................................ ........ 107 6 EFFECT OF THERMAL STRESS ON RELIABILITY ................................ ............ 127 6.1 Importance of Understanding the Thermal Expos ure to HEMTs ..................... 127 6.2 Experimental ................................ ................................ ................................ ... 127 6.2.1 Iso chronal Anneals ................................ ................................ ............... 127 6.2.2 Iso thermal Anneals ................................ ................................ ............... 128 6.3 Results and Discussion ................................ ................................ ................... 129 6.3.1 Iso chronal Anneals on SiC HEMTs ................................ ...................... 129 6.3.2 Iso chronal Anneals on Si(111) HEMTs ................................ ................ 130

PAGE 8

8 6.3.3 Iso thermal Anneals ................................ ................................ ............... 135 6.3.4 E ffect on I G V T I D and Contact Resistance ................................ .......... 136 6.3.5 Simulation of Metal Penetration Effect on V T and I D .............................. 138 6.3.6 Capacitance Voltage Results ................................ ............................. 140 6.4 Summary ................................ ................................ ................................ ........ 142 7 CONCLUSIONS AND FUTURE WORK ................................ ............................... 167 APPENDIX A FLOOPS SCRIPTS FOR GATE STRUCTURES ................................ .................. 171 B DERIVATION OF GATE METAL PENETRATION DEFECT SIZE FOR SIMULATION ................................ ................................ ................................ ........ 182 LIST OF REFERENCES ................................ ................................ ............................. 185 BIOGRAPHICAL SKETCH ................................ ................................ .......................... 192

PAGE 9

9 LIST OF TABLES Table page 6 1 Electr ical data from 0.5 h anneals on Si(111) AlGaN/GaN HEMTs .................. 143

PAGE 10

10 LIST OF FIGURES Figure page 2 1 Schematic of wurtzite GaN. A) Ga face. B) N face. ................................ ............ 50 2 2 XTEM micrograph showing the typical structure of an AlGaN/GaN HEMT. ........ 50 3 1 Schematic of the tube furnace set up used for therma l annealing. ..................... 64 3 2 SEM and ion beam micrographs showing the FIB procedures to lift out a XTEM sample of an AlGaN/GaN HEMT. ................................ ............................ 65 3 3 SEM and ion beam micrographs showing the FIB procedures to fabricate an APT tip of an AlGaN/GaN HEMT. ................................ ................................ ....... 66 3 4 Schematic of the electropolishing system used to sharpen the W wires into sharp tips. ................................ ................................ ................................ ........... 67 3 5 Diagram of the setup of the TLM measurement technique. ................................ 68 3 6 Schematic of the internal configuration of a TEM ................................ ............... 69 4 1 HAADF STEM images of the gate region of AlGaN/GaN HEMTs showing an increase in the interfacial lay er thickness from left to right ................................ 83 4 2 Imago LEAP reconstructions of the gate/AlGaN interface showing the Ni/Au gate metal stack and AlGaN and GaN epilayers.. ................................ .............. 84 4 3 Mass histories from field evaporation of the gate/AlG aN region of HEMTs using the Imago LEAP 3000X Si system.. ................................ .......................... 85 4 4 Partial M/C spectra collected from APT analysis of the gate region of an AlGaN/GaN HEMT structure on Si(111). Partial spectra are used to deduce a lack of Ga oxide in the interfacial oxide. ................................ .......................... 86 4 5 1D concentration profiles from the gate/AlGaN interface region of an AlGaN/GaN HEMT on Si(111) ................................ ................................ ............ 87 4 6 BSE micrographs o f APT tips from AlGaN/GaN HEMTs ................................ .... 88 4 7 A XTEM image and BSE micrographs of the orientation of the gate and epilayers of an AlGaN/G aN HEMT.. ................................ ................................ ... 89 4 8 A partial APT reconstruction showing part of the AlGaN epilayer and the GaN layer below it.. ................................ ................................ ................................ ..... 90 4 9 XTEM image and EELS chemical maps of the gate/AlGaN interface regi on from a Ni gate on SiC HEMT ................................ ................................ .............. 91

PAGE 11

11 5 1 Low magnification HAADF STEM image of a cross section of an unstressed Ni gate AFRL AlGaN/GaN HEMT o n SiC with the source (S), gate (G), and drain (D) contacts labeled. ................................ ................................ ............... 109 5 2 Measured normalized I D max and I G degradation HEMTs when stressed from V GS = 5 or 10 V to 42V at 1 V/min and V DS maintained at 5 V throughout stressing. ................................ .. 110 5 3 High magnification HAADF STEM images of the gate region of AlGaN/GaN HEMTs showing the distinct Ni/Au g ate metal stack layer s and AlGaN and GaN epilayers. ................................ ................................ ................................ 111 5 4 Measured normalized I D max degradation HEMTs when stressed from V GS = 5 or 10 V to 4 2V at 1 V/min while V DS maintained at either 0, 5, 10, or 15 V throughout stressing. ... 112 5 5 Measured normalized I D max and I G degradation AlGaN/GaN HEMT str essed from V GS = 10 V to 42V at 1 V/min while V DS was maintained at 0 V throughout stressing. .......................... 113 5 6 Measured normalized I D max and I G degradation AlGaN/G aN HEMT stressed from V GS = 5 V to 42V at 1 V/min while V DS was maintained at 10 V throughout stressing. ........................ 114 5 7 Measured normalized I D max and I G degrada tion AlGaN/GaN HEMT stressed from V GS = 10 V to 42V at 1 V/min while V DS is maintained at 15 V throughout stressing. ............................ 115 5 8 High magnification HAADF STEM images of the gate region of AlGaN /GaN HEMTs showing the distinct Ni/Au gate metal stack layers and AlGaN and GaN epilayers. ................................ ................................ ................................ .. 116 5 9 High magnification HAADF STEM images of the gate region of AlGaN/GaN HEMTs and their correspo nding E DS line scans ................................ ............. 117 5 10 High magnification BF overlaid with EDS maps ................................ ................................ .................. 118 5 11 High magnification HAADF defect overlaid with EDS maps. ................................ ................................ ........ 119 5 12 High magnification BF efect overlaid with EELS maps ................................ ................................ ................. 120 5 13 High magnification HAADF d efect overlaid with EELS maps. ................................ ................................ ...... 121

PAGE 12

12 5 14 Simulation showing the iso contour lines of the magnitude of the vertical and stress state of V DS = 5 V and V GS = 42V. ................................ ........................ 122 5 15 Simulation showing the iso contour lines of the magnitude of the vertical and lateral components of the electric field from Ni gat e HEMTs structures by Gao et al ................................ ................................ ................................ .......... 123 5 16 Representative BF TEM images show how the air gaps in the SiN x passivation lead to ambient and the channels connec ting the gate edges to the gaps ................................ ................................ ................................ ........... 124 5 17 Measured I G representativ e curves of stressed AFRL AlGaN/GaN HEMTs on SiC demonstrating the before, middle, and after I G jump characteristics.. ........ 125 5 18 High magnification HAADF STEM images of the gate region of AFRL AlGaN/GaN HEMTs on SiC showing the distinct Ni/Au gate metal stack layer s and AlGaN and GaN epilayers. ................................ .............................. 126 6 1 Low magnification XTEM images of the gate region of AFRL AlGaN/GaN HEMTs on SiC s howing the distinct Ni/Au gate metal stack layer s and AlGaN and GaN epilayers ................................ ................................ ........................... 144 6 2 Low magnification HAADF STEM images of the gate region of AFRL AlGaN/GaN HEMTs on SiC showing the distinc t Ni/Au gate metal stack layer s and AlGaN and GaN epilayers. ................................ .............................. 144 6 3 PSEM images of the top of the AlGaN layer ................................ .................... 145 6 4 PSEM ima ges of the top of the AlGaN layer of 1 m gate lengt h AFRL AlGaN/GaN HEMTs on SiC ................................ ................................ ............. 146 6 5 High magnification PSEM image of the top of the AlGaN layer of an AFRL AlGaN/GaN HEMT on SiC showing the hexagonal symmetry of the pits indicated by white arrows. ................................ ................................ ................ 147 6 6 HAADF STEM images of the gate region and epilayers of Ni gate AFRL AlGaN/GaN HEMTs on Si(111) showing the gate metal penet ration in dislocations.. ................................ ................................ ................................ ..... 148 6 7 High magnification HAADF STEM images of the gate/AlGaN interface of AFRL HEMTs on Si(111) showing the gate metal penetration depth into the AlGaN epilayer afte r a 500C anneal. ................................ .............................. 149 6 8 XTEM image of gate metal penetration and EDS chemical composition analysis ................................ ................................ ................................ ........... 150

PAGE 13

13 6 9 HAADF STEM images of only single metal coated AlGaN/GaN epilayers after a 500 C anneal for 0.5 h ................................ ................................ .......... 151 6 10 High magnification HAADF STEM images of the gate/AlGaN interface of HEMTs on Si(111) after a 0.5 h anne al ................................ ........................... 152 6 11 Plot of the rate of metal penetration vs. temperature for 0.5 h iso chronal anneals for HEMTs on Si(111) from 300C to 500C ................................ ....... 153 6 12 High magnification HAADF STEM images showing the gate metal pen etration after a 300C anneal ................................ ................................ ..... 154 6 13 Plot of the gate metal penetration distance vs. duration of anneal at 300C .... 155 6 14 Normalized I G plot showing anneals at 300C, 400C, 450C 500C, and 600 C for 0.5 h. ................................ ................................ ................................ ..... 156 6 15 Change in I G af ter a 0.5 h anneal at 300C indicating that the I G is always lower than the unstressed condition. ................................ ................................ 157 6 16 Change in I G after a 0.5 h anneal at 400C indicating that the I G is always smaller t han the unstressed condition when V GS < 0.7 V. ................................ 158 6 17 Change in I G after a 0.5 h anneal at 450C. ................................ ...................... 159 6 18 Change in I G after a 0.5 h anneal at 500C. ................................ ...................... 159 6 19 Change in I G after a 0.5 h anneal at 600C indicating that the I G is always larger than the unstressed condition. ................................ ................................ 160 6 20 Representative V T shift in an AlGaN/GaN HEMT annealed at 400C for 0.5 h. 160 6 21 Normalized I D from the 400 C anneal for 0.5 h indicating that the I D has increa sed from the stress. ................................ ................................ ................ 161 6 22 Normalized I G for HEMTs exposed to a 300C anneal for up to 400 h. The figure indicates that all HEMTs exhibited lower I G below V GS = 0.7 V. .............. 162 6 23 Shift in V T over time for HEMTs exposed to a 300C anneal up to 400 h. ........ 163 6 24 Normalized I D for HEMTs exposed to a 300C anneal for up to 400 h. The figure indicates that I D decreased with time. ................................ ..................... 164 6 25 Gate structures used in simulations to determine the effect of ga te metal penetration on HEMTs l. ................................ ................................ .................... 165 6 26 Representative C V curves from an AFRL HEMT on Si(111) before and after a 400C anneal for 0.5 h. ................................ ................................ .................. 166

PAGE 14

14 LIST OF ABBREVIATIONS 1D One dimensional 2DEG Two dimensional electron gas 3D Three dimensional ALD Atomic layer deposition APT Atom probe tomography BF Bright field BSE Backscattered electron C V Capacitance Voltage EDS Energy dispersive X ray spectroscopy E F Fermi energy level E G Band gap EELS Electron energy loss spectroscop y FIB Focus ed ion beam FIM Field ion microscopy G m Transconductance HAADF High angle annular dark field HEMT High electron mobility transistor I DS Drain current I DSS Saturated drain current I G Gate leakage current LEAP Local electron atom probe M/C Mass to charge MBE Molecular beam epitaxy MESFET Metal semiconductor field effect transistor

PAGE 15

15 MOCVD Metal organic chemical vapor deposition PSEM Plan view scanning electron microscopy ROI Region of interest SE Secondary electron SEM Scanning electron microscopy S HB Schottky barrier height STEM Scanning transmission electron microscopy TEM Transmission electron microscopy V DG Gate to drain voltage V DS Drain to source voltage V GS Gate to source voltage V T Threshold voltage XSEM Cross sectional scanning electron micr oscopy XTEM Cross sectional transmission electron microscopy Z Atomic number

PAGE 16

16 Abstract of Dissertation Presented to the Graduate School of the University of Florida in Partial Fulfillment of the Requirements for the Degree of Doctor of Philosophy IMPACT OF ELECTRICALLY AND THERMALLY INDUCED PHYSICAL DEFECTS ON THE RELIABILITY OF ALGAN/GAN HIGH ELECTRON MOBILITY TRANSISTORS By Monta Raymond Holzworth, Jr. August 2013 Chair: Kevin S. Jones Major: Materials Science and Engineering AlGa N/GaN high electron mobility transistors are unique for their combination of high temperature, high power, and high frequency applications. Compare d to Si, Ge, and compound semiconductors such as GaAS and InP, AlGaN/GaN transistors outclass the current te chnology due to their sup erior combination of high break down voltage and high frequency performance. These characteristics arise from structural and electrical properties inherent to the AlGaN/GaN heterojunction which have enabled AlGaN/GaN transistors us age in important military and civilian applications such as microwave and millimeter technology, RADAR systems, and as high current and voltage switches in utility grid systems As the technology continues to improve due to increased materials quality and device advancements future applications wi ll require AlGaN/GaN transistor usage under even higher voltage s and temperature s Therefore, the effects of these stresses need to be investigated in order improve device performance and reliability When stres s conditions are applied in combination, device failure is accelerated. H owever, future applications may require electrical or thermal stress to be effectively applied separately. Therefore, it is important to understand how each factor contributes

PAGE 17

17 indiv idually to the reliability and failure mechanism s in transistors to determine the actual workin g device life time in its operating environment Th e following research employs structural, chemical, and electrical device c haracterization paired with simulat ion in order to develop structure property relationship s between defects and device performance Here, for the first time, as grown gate interfacial layers were characterized using atom probe tomography and are composed of two distinct oxide layers NiO x and AlO x K nowing the composition permits elimination or reduction of the layers through etching or processing modifications Furthermore, u sing off state reverse bias electrical stress, Ni gate metal reactions with AlGaN epilayer s emulate the shape and size of the electric field contours between 5 6 MV/cm ; an advancement from the previous understanding that defects form only at the peak electric field. Finally application of only thermal stress is shown for the first time to cause gat e metal penetra tion into threading dis locations This penetration combined with Ni Au gate metal interdiffusion causes a negative shif t in threshold voltage, a n increase in drain current, and only above 400C an increase in gate leakage.

PAGE 18

18 CHAPTER 1 INTRODUCTION Motivat ion AlGaN/GaN High Electron Mobility Transistors (HEMTs) exhibit excellent high frequency, high power, and high temperature performance and have the potential to replace Si and GaAs based transistors for a number of applications. 1 8 Specifically, applicat ions for these transistors include both civilian and military functions such as microwave applications broadband wireless access, cellular infrastructure, utility grid application s, optoelectronic devices, RADAR communications, and space application s. A lGaN/GaN HEMTs are remar kable for both their large breakdown voltages and high operating frequencies. The combination of these two features makes them superior to similar Si, S iGe, GaAs, and InP transistors. Additionally, AlGaN/GaN HEMTs p ossess large ba nd gaps, high electron saturation velocities, and a high elec tron sheet carrier density. 9 10 Another advantage of AlGaN/ GaN HEMTs is the ability to fabricate these transistors on a ra nge of different substrates including SiC sapphire and Si. This permit s tuning the transis tors to specific applications s uch as u sing AlGaN/GaN HEMTs on SiC for a larger thermal budget when high temperatur e applications are considered. 11 D espite the inhe rent material advantages and wide range of applications, AlGaN/GaN HEMT s are plagued by a multitude of failure mechanisms ranging from electrical to physical. 12 2 1 Many of these mechanisms are not well understood and are just beginning to be explored due to improvement s in device material quality and design Addi tionally, spe cific components of the transistors are prone to failure such as the gate region of the device. Normally, accelerated testing using harsh multiple

PAGE 19

19 stressing methods is applied to HEMTs in order to induce failure and analyze the root cause. However, the application of these multiple simultaneous stresses can overly accelerate f ailure and therefore not represent the true failure me chanisms. However, in future applications, these stresses may be effectively applied one at a time. For example, a transisto r may be stored in a high temperature ambient, but operate during off state conditions for the majority of its lifet ime resulting in its life time limited by failure induced by thermal stress. Therefore, it is important to understand how each of these fa ctors contributes individually to reliability issues and failure mechanisms in transistors in order to determine the actual working life time of the transistor in its operating conditions. Thus, in this work electrica l and thermal stress are studied inde pendently in order to understand how each stress changes the device structure, electrical properties, and device reliability. Objectives of Research The goal of this work is to investigate the formation of physical defects in AlGaN/GaN HEMTs and their eff ect on H EMT reliability. These defects could be as grown in the transistor due to the growth and fabrication process, or they may arise due to accelerated stress testing. Due to the complex nature of defect formation, the variables influencing defect for mation are isolated and broken up by their electrical and thermal components. Dissertation Outline The purpose of this work is to identify, characterize and model failure mechanisms in AlGaN/GaN HEMTs. Therefore, the properties, structure, and device p hysics of AlGaN/GaN HEMTs will be covered in Chapter 2 along with summary of physical defect mechanisms that have been previously identified. The material

PAGE 20

20 processing for this work will be covered in Chapter 3 in addition to the various methods of sample c haracterization used to analyze HEMTs Structural and c hemical characterization of as grown interfacial layers that are known to affect HEMT reliability is presented in Chapter 4 along with the difficulties in procuring accurate nano scale data Defects generated by the application of large electric fields and their e ffect on HEMT reliability will be presented in Chapter 5. Finally d efects created by thermal proc esses and their effect on HEMT reliability will be covered in Chapter 6

PAGE 21

21 CHAPTER 2 BACKGROUN D In the following sections of this chapter, background information on AlGaN/GaN HEMTs will be covered. First the material and electronic properties of GaN and AlGaN are presented to provide a n explanation on why these materials were chosen to make transi stors. Next, information summarizing the structure and fabrication of devices made from GaN and AlGaN is presented to clarify the device structure and properties exhibited by the devices that are not present in its constituent binary semiconductors A su mmary of the device physics and theory of working components o f an AlGaN/GaN HEMT will follow to explain how the individual parts of the transistors operate. Finally a review of the current understanding and research on failure mechanisms and reported ph ysical degradation of AlGaN/GaN HEMTs is presented. 2.1 Properties of GaN and AlGaN Together GaN and AlN make up the building blocks of AlGaN/GaN transistors. When GaN is desired as a semiconductor layer, GaN is grown as a binary compound semiconductor. However, the growth of the tertiary compound semiconductor AlGaN requires a combination of GaN and AlN When neglecting the gaseous precursors, th is relationship can simplistically be expressed as: (x) AlN + (1 x) GaN = Al x Ga 1 x N (2 1) where x is the molar fraction of AlN; however, it is noted that the specific chemical reaction that forms AlGaN will vary with the growth technique used in synthesis but this general ratio of AlN to GaN in the AlGaN would apply. Due to the large difference in band gap E G between GaN at 3.4 eV and AlN at 6.2 eV AlGaN has a tunable E G

PAGE 22

22 between these values based upon its Al composition. The E G for AlGaN can be expressed (if bowing is neglected) as: E G (eV) = 3.4 + 2.8 x (2 2) w here x is the molar fraction of AlN. G good thermal conductivity, and high thermal stability, they are suitable materials for high temperature applications. In addition to it s material proper ties, GaN also has unique electrical properties. For example, GaN has a high breakdown voltage ~5 x 10 V cm 1 a bulk electron mobility of ~1000 cm 2 V 1 s 1 and a saturation velocity > 3 x 10 7 cm s 1 9 10 These properties make GaN based devices excellent choices for high electric field and frequency applications. GaN and consequently A lGaN possess two allotropes. They can be grown with a cubic, specifically zin c blende crystal structure or they can be grown with a hexagonal, wurtzite structure. For this work only hexagonal, wurtzite GaN was used. Additionally, for GaN the wurtzite structure is more commonly used in order to better lattice match it to Si (111 ) which w hen u sed as a substrate has a hexagonal symmetry, too. GaN possesses two other struc tur al features result ing in additional physical and electric al properties. First, in wurtzite GaN one layer of the crystal is composed of all cations and the next layer is composed of all anions. This arrangement results in a repeating, alternating structu re of bonded ions. Due to this alternating structure when GaN growth terminates, its last monolayer will be either all Ga or all N. This gives GaN an orientation dependence that is referenced by either Ga face for a terminal monolayer of all Ga or N fa ce for a terminal monolayer of all N. The structures of Ga face and N face GaN are shown in Figure 2 1 A and Figure 2 2 B respectively. Second,

PAGE 23

23 wurtzite GaN is non centrosymmetric and therefore lacks an inversion center (center of symmetry) This lack of symmetry the alternating layers of ions from an idea l c/a axis ratio gives wurtzite GaN and AlGaN a spontaneous polarization, piezoelectric ity and piezoelectric polarization. For wurtzite GaN and AlGaN, the spontaneous pola r ization dipole forms along the <0001> direction. Like the E G of AlGaN, the spo ntaneous polarization is depende nt upon the molar fraction of Al due to the diffe rence in polarizations between GaN at 0.029 C m 2 and AlN at 0.081 C m 2 22 Piezoelectricity is the ability of a crystal to form an internal electric field when strain is applied, or inversely if an electric field is applied to the crystal it will undergo a strain. This additional strain can be found by: ij = S ijkl kl + d kij E k (2 3 ) where S ijkl kl is the stress tensor, d kij is the piezoelectric modulus, and E k is the electric field at a specif ic, constant temperature 23 Alternatively, due to the mechanical relationship between stress and strain, the addit ional strain can instead be calculated as the additional stress added to the device and is given by: ij = C ijkl kl e kij E k (2 4 ) where C ijkl kl is the strain tensor, and e kij is the piezoelectric c oefficient tenso r 2 4 25 Both d kij and e kij are tensors that describe the piezoelectric properties of GaN. The difference between them is their derivation. d kij is derived as the change in electric charge density displacement over the change in temperature at constant el ectric field However, e kij is derived as the change in electric charge density displacement over the change in strain at constant electric field. These

PAGE 24

24 piezoelectric co nstants are a third rank tensor. D symmetry there are only three independent nonzero coefficients: d 31 = d 3 2 d 33 and d 15 = d 24 xx yy zz ) due to the vertical electrical field in the HEMT yz xz ) due to the hor izontal field 2 3,26 This is an important property of GaN and AlGaN as future sections (Section 2.4 .6 and Chapter 5) of this work will explain how the strain caused by the application of large electric fields in AlGaN/GaN HEMTs can cause early failure of de vices and the formation of defects. lattice constant than GaN. Thus, when AlGaN is pseudomorphically grown on a GaN layer below the critical thickness, it is put into tension in order to match the underlying GaN. This strain in the AlGaN layer creates a piezoelectrically induced polarization in addition to the normal spontaneous polarization already present in AlGaN. 2.2 Transistor Fabrication Due to the great er expense, lower q uality, and smaller size of GaN substrates, AlGaN/GaN HEMTs are grown by heteroepitaxy on a variety of substrates. Commonly used substrates are sapphire (Al 2 O 3 ), Si (111), and SiC each with its own advantages and disadvantages. Sapphire has a lattice an d thermal expansion mismatch with GaN of 1 6% and 39%, respectively 27 Additionally, sapphire has a relatively po or thermal conductivity 28 29 Meanwhile Si and GaN have similar thermal conductivities, but the lattice and thermal expansion mismatch between t hem is large at 16% and 56% respectively 29 30 Although using a Si substrate results in the largest lattice mismatches, some of its advanta ges are the low cost, commercial availability, and large wafer size. On the other hand, SiC has excellent thermal c onductivity and the smallest lattice and

PAGE 25

25 thermal expansion mismatch es of 3.5% and 3.2% respe ctively 27 29 D ue to the lattice mismatch es between the substrate and GaN many layers typically called nucleation and transition layers are needed in order to r educe the strain to permit GaN growth on the substrates. These layer s can be seen in the cross section al transmission electron microscopy (XTEM) micrograph of a typical structure for an Al GaN/GaN HEMT is shown in Figure 2 2 Layer growth in AlGaN/GaN HE MTs uses either m olecular beam epitaxy (MBE) or metal organic chemical vapor deposition (MOCVD) The nucleation layer is the first layer to be deposited; it helps the transition layers and therefore the GaN and AlGaN epilayers adhere to the substrate. Th e chemistry of the nucleation layer varies by substrate. Following the nucleation layer, the next set of layers are the transition layers. Because of the large lattice mismatch between the substrate/nucleation layers and GaN, misfit and threading disloca tions can form and propagate through the material and into the epilayers. One function of the transition layers is to prevent this from occurring or at least mitigate some of the threading dislocation density by decreasing the mismatch between the substra t e/nucleation layers and the GaN 9,29,31 The transition layer helps by grading the lattice parameter difference between the two instead of depositing GaN directly on the substrate where the change in lattice parameter would be immediate. There are two com monly used transition layer schemes in HEMT fabrication. First, there is a graded layer. Here for example, the nucleation layer may be some type of AlN layer. Then the composition will be graded from Al x G a 1 x N to GaN where x =1 at the nucleation layer and x = 0 at the beginning of the GaN layer over a distance o f a few microns. Second, there is a n abrupt multi layer scheme. Here there

PAGE 26

26 may be 3 5 transition layers each having a different, constant composition. However, the compositions between th e layers change from more Al rich to Ga rich as the layers progress further from the substrate/nucleation layer. In this work, the abrupt transition layers were used in the AlGaN/GaN HEMTs. A nother benefit of using the transition layers is to prevent res idual th ermal strain cracking 32 Following the transition layer is the growth of the GaN buffer and AlGaN layers Typically the GaN buffer is Fe doped. This helps reduce the short channel effects and confines the vertical component of the electric field to the AlGaN epilayer. AlGaN has a smaller lattice parameter than GaN. Thus, w hen AlGaN is pseudomorphically grown on GaN below the critical thickness, a permanent tensile strain develo ps within the layer 33 One advantage to using a thinner AlGaN layer is better radio fr equency (RF) reliability 34 This is attributed to the reduction in strain by decreasing the AlGaN layer thickness relative to the critical thickness and lowering the Al composition 35 Occasionally, the AlGaN/GaN heterojunction is capped with a thin (~1 2 nm) GaN layer. Capping the AlGaN e pilayer with GaN yields a few of benefits. First, it decreases the number of seeds from which a crack can initiate by smoothing the surface. This could increase device reliability under high bias co nditions and help p rotect it from damage due to piezoelectric strain Secondly it decreases the likelihood of an oxide developing on the AlGaN/GaN surface The oxide c ould form an interfacial layer between the AlGaN and gate affecting the HEMT electric al characteristics Thirdly, i t reduces the oxygen trap density at the AlGaN/gate interface 36 Lastly, it reduces the gate leakage current I G Gate leakage current reduction could be achieved in two ways. One method is by increasing the effective Schot tky barrier height between

PAGE 27

27 the gate and AlGaN using the inhe rent piezoelectric charge 37 The second method is by reducing the flow of electrons into threading dislocations by reducing the magnitude of the vertical electric field 38 This lowers the I G by de creasing the Frenkel Poole emission from surface and near surface states to threading dislocations. Following layer growth, the metal contacts are deposited. T here a re two different types of contacts on the HEMT. The sou rce and drain contacts are ohmic a nd are deposited first and followed by a short, high temperature anneal However, the gate contact is Schottky and does not experience a thermal anneal While many different gate metal stack combinations have been investigated for the best thermal reliab ility, the two most commonly used are Pt/Au and Ni/Au. In this work, the gate stacks used are Ni/Au. A passivation layer is used to cover the gate contact and top surface of the AlGaN epilayer. The most commonly used passivation layer is MO CVD SiN x alt hough atomic layer deposition ( ALD ) Al 2 O 3 is being used more frequently in research applications The deposition of the SiN x passivation layer also decreases the number of surface traps which improves HEMT performance by incr easing the drain current I DS 39 If it is incorporated into the device design, typically the last f eature that is fabricated is the field plate. The inclusion of a field plate in the device structure helps to mitigate the peak electric field by spreading the field out further along th e access region This improves performance for microwave applications by reducing t he I G 40 Field plates can be connected to the source or gate contacts and consist of metals that match those found in the contact stacks making one extended, continuous con tact.

PAGE 28

28 2.3 Device Physics A HEMT is made up of multiple components that help determine its output characteristics. Two of the most important features are the contacts. The gate contact which switches the device from off to on mode operation is based o n a Schottky contact which permits current to flow in only one direction. Additionally, the source and drain contacts are ohmic contacts which allow easy current trans port from the conductive two dimensional electron gas ( 2DEG ) through the ohmic contacts and out through the device These features are described in more detail in the following sections. 2.3.1 Schottky Contacts Schottky contacts are defined by their rectifying properties where almost no current flows until a specific forward bias O nce this bias is achieved high conduction occurs and the current flow b ecomes exponential in behavior. It is possible for large current flow to occur under reverse bias; however, this only occurs when the reverse bias become s too large This causes breakdown r e sulting in a large amount of current. When a Schottky contact is required for an n type semiconductor, a metal with a larger work function than the semiconductor work function is used. This creates a barrier between the gate and semiconductor conduction bands. However, it is noted that it is possible for electrons to tunnel through the barrier. In theory the properties of contacts depend upon the Schottky barrier height ( SBH ) Although metal work functions may be consulted when choosing a chemistry of a contact stack, typically interface and gap states inherent to the semiconductor or caused by the transition from the semiconductor to the metal cause the Fermi level E F to b e pinned. Additionally, surface contamination or reactions between the

PAGE 29

29 semicon ductor and metal can influence the electronic properties of the interface making the contact behave non uniform. Due to t hese factors the behavior of the contacts typically does not match the predicted behavior from their work functions Thus, contact behavior is difficult to predict a priori without first experimentally measuring the contact properties. 2.3.2 Ohmic Contacts O hmic contact s are defined by their linear relationship between current and voltage under both forward and reverse bias which ag In theory, ohmic contacts can be formed on a n type semiconductor, by using a metal with a smaller work function than the semiconductor work function. Similar to the Schottky contacts due to Fermi level pinning, gate metal and semic onductor reactions, or contamination, c ontact behavior deviates from the predictive t heory. Another issue is scarcity of metals with low work functions. One way around this problem is to highly dope the semiconductor region of the contact A highly dope d region would narrow the depletion width at the metal semiconductor interface which would allow e le ctrons to more easily tunnel through the barrier Another method is to use a thin layer of a small E G s emiconductor between the original metal s emiconducto r interface Finally one last method for forming an ohmic contact is to use a metal stack that reacts with the semiconductor in order to penetrate the metal to the conductive channel. 2.3.3 Device Operation Like all field effect transistors, HEMTs have a threshold voltage V T which (no current flow). In AlGaN/GaN HEMTs the V T is determined by charges accumulated at the interfaces between the gate metal and AlGaN and the AlGaN and GaN. Here, the V T is given by:

PAGE 30

30 V T = V BI + E F C (t AlGaN PZ AlGaN ) (2 5 ) w here V BI is the built in potential, E F is the Fermi energy of the AlGaN, C is the conduction band offset between the AlGaN and GaN t AlGaN is the thickness of the AlGaN layer PZ is the combined polarization charge due to spontaneous polarization and piezoelectric polarization at the AlGaN/GaN interface and AlGaN is the permittivity of the AlGaN These charges are modulated and controlled by the change in gate to sour ce voltage, V GS Here, for the off state condition, the V GS is not greater than the V T Under this condition, the 2DEG below the gate contact is depleted of the electron charge carriers This prevents current from flowing between the source and drain. However, as V GS increases above V T and a voltage is applied to the drain contact, V DS the 2DEG begins to fill and the current increases linearly. Following the V GS > V T condition, as the V DS continues to increase such that V DS > V DS V T a saturation of the drain current occurs. Under this condition, the application of additional V DS does not result in any additional current flow. 2.4 Reliability and Degradation Issues in AlGaN/GaN HEMTs Despite the useful electrical and materials properties that give AlGaN/GaN HEMTs a large spread of applications, there are many unresolved reliability issues that re main with these transistors and limit their functionality. While there are electrical degradation of HEMTs. Thus, only a cursory review of electrical defect s will be covered in contrast to a more in depth review of physical defects in the upcoming sections

PAGE 31

31 2.4.1 Surface Traps The creation of surface traps is a common occurrence in AlGaN/G aN HEMTs. They are measured electrically using gate lag measurements. Thus, changing the device surface using different processing steps such as passivation or chemical/plasma processing and then measur ing the gate lag correlated the electrical measureme nts to how the process affected the surface traps. Surface traps act by trapping electrons in states near the surface of the device. These trapped electrons then accumulate and deplete the 2DEG. Due to this depletion, electron flow is limited which caus es the I DS to decrease. By reducing the number of available trap sites, the depletion of the 2DEG is reduced which causes the normalized I DS to be lowered a smaller amount than the original. Thus, if a process causes an increase in the normalization I DS then the number of surface traps was most likely reduced. 2.4.2 Buffer Traps Due to the many different layers used in AlGaN/GaN HEMTs, there are many interfaces that have the possibility of acting as traps. In particular, traps at the GaN buffer interf ace play a role in device reliability. Here, the t raps in the GaN buffer layer are respons ible for current collapse in AlGaN/GaN HEMT s This was first reported by Khan et al. when a 20V drain stress was applied and caused a severe decrease in I DS 41 Addi tionally, Binari et al. reported current collapse in GaN metal semiconductor field effe ct transistors (MESFETs) 42 The I DS was measured with and without photo illumination during V DS stressing and the dependence on the I DS recovery with wavelength was note d The smaller wavelength (higher energy) photo illumination resulted in the greatest recovery Additionally, a s the V DS was increased the current

PAGE 32

32 collapse worsened This demonstrated that the smaller wavelength photo illumination release d more electro ns from their trap sites. Furthermore, buffer trapping can be measured using drain lag measurements where the time dependence of the recovery from current collapse by thermal emission of trapped carriers and the temporal response of the current collapse is measured 43 In d rain lag measurements, a voltage pulse is used and then the I DS is measured before and after the pulse. The ratio of the measured I DS is taken and the closer it is to unity the less buffer trapping there is in t he HEMTs 39 2.4.3 Hot Elect ron Effects Th e creation of hot electrons is another reliability issue in AlGaN/GaN HEMT s due to these hot electrons generating both traps and physical damage in the device. They are formed due to the high electric field (~several MV/cm) present during op eration which accelerates the electron s so that they have a large velocity and therefore large energy. Due to their large kinetic energy, they can penetrate features throughout the device and be trapped at many different locations such as on the device su rface, in the AlGa N la yer, or in the GaN buffer 11 Hot electrons can also create defect traps from impact damage. Jha et al. have shown that interface trap density at the AlGaN/GaN interface increases sig nific antly from hot electrons 44 Additionally, trap s created by hot electrons can increase the depletion region which results in a decrease in I DS Additionally, Sozza et al. showed that in AlGaN/GaN HEMTs subjected to a 3000 h on state electrical stress that surface traps were created which led to a decr ease in I DS and an increase in drain resistance 45 Furthermore, they noticed for HEMTs stressed with off state bias conditions that minimal change in the I DS and resistance occurred. These changes were associated with a slight increase in trapping It i s noted that the

PAGE 33

33 majority of devices in this work were subjected to off state or no electrical stress and thus hot electron effects have been omitted in the analysis. 2.4.4 Gate Contact Degradatio n Contact degradation in HEMTs can occur through multiple mec hanisms. Metals can interdiffuse, form intermetallic phases, or they can react with the semiconductor. These issues can change the electrical properties of the contacts and therefore affect the device reliability. One of the most common reasons for cont act degradation is the amount of thermal energy a contact receives. Another degradation mechanism for contacts is gate sinking which is an important degradation mechanism in GaAs transistors. However, it has not been directly reported in the AlGaN/GaN HE MT literature Occasionally, there are reports of some reaction s between the gate metal and semiconductor epilayers, but this has only occurred when it appears to be the result of dislocations Gate sinking was investigated by Chou et al. in AlGaN/GaN HEMTs using accelerated life tests. A temp erature step stress starting at 150C and ending at 240C with 48 h intervals and 15C steps was used to stress devices. No gate sinking was observed 31 Another study subjected devices to a temperature stress tes t and a combined thermal electrical stress test. During the temperature stress test, devices we re heated to 400C for 1000 h The combined stress test occurred at 350C and had a continuously injected current using the biasing conditions of V DS = 20 V an d V G S = 0 V. Again, n o diffusion of the gate metals into th e semiconductor occurred 46 One type of gate contact degradation is the presence of a native interfacial layer between the gate and AlGaN layer This can result in a decrease i n the saturated dra in current, I DSS Studies have shown that when HEMT s are operated at high

PAGE 34

34 temperatures, the interfacial layer is modified (thickness decreases) and may dissolve which causes a positive increase in the SBH. This permanent shift in SBH results in the decre ase in I DSS 47 48 Additionally, H EMTs exposed only to a thermal anneal also exhibited this decrease in interfacial layer thickness. Furthermore, these devices exhibited 50% less degradation compared to unannealed devices. Marcon et al. performed long term thermal test on Ni/Au gate AlGaN/GaN HEMT s for 1000 h at 325C. It was noted that after a short exposure to elevated temperatures that device performance improved. H owever, after 384 h of exposure to 325C, the device performance began to degrade. The I DSS increased, the V T shifted more negatively by 0.4 V, and I G and I D increased 1 to 2 orders of magnitude, respectively, with no saturation occurring. The degradation was attributed to the interdiffusion of the Ni and Au layers in the gate Schottky cont act by imaging using scanning electron microscopy (SEM). 8 2.4.5 Source and Drain Contact Degradation Stable ohmic contacts are important feature s for the HEMT operation. If the ohmic contacts degrade, then the electrical outputs of the device also suffer because less I DS is conduct ed through the contact. O hmic c ontacts can degrade chemically or morphologically. When contacts are subjected to high temperature anneals, the met al stack breaks down into different phases that are combinations of the original metals present 39 However, these structural and chemical change s may not be detrimental to the electrical properties of the contact. For example, t he contact resistance of an ohmic stack can be compared to its annealed temperature. This results in a mini mum contact resistance at a specific temperature. Knowing this processing temperature is important for producing ohmic contacts with the best electrical properties. However, subjecting

PAGE 35

35 the contact to extra thermal energy would increase the cont act resist ance and reduce, adversely affect the I DS For example, Piazza et al. showed that for Ti/Al/Ni/Au ohmic contacts stored for 2000 h above 290C that the contact resistance increased. Additionally, the contact surface roughened due to the growth of Au ric h grains 49 Within the contact it was found that Au interdiffused among the ohmic metal stack, and another issue involved the out diffusion of Ga from the epilayers and into the contact. Furth er more, similar degradation was found to occur in ohmic contact s when an electrical stress was performed to make the internal device tempe rature equivalent to the 2000 h stored thermal tests. In addition to chemical degradation, morphology changes c an occur When the contacts are subjected to high temperature anneals their surface can roughen and their rigid edges degrade and become curved. It has been observed that the stability of contact morphology at high temperatures can be improved by adding W or WSi x to the ohmic stack 50 A gain, due to the self heating during on state operation, monitoring the thermal budget of the HEMT s is an important component of device reliability due to the negative electrical and structural changes possible from the long term application of thermal energy. 2.4.6 Inverse Piezoelectric Ef fect The inverse piezoelectric effect is a concern in AlGaN/GaN HEMTs because of the large piezo coefficients in nitrides and the large voltages used during operation and reliability testing Typically t he largest electric field occurs at the drain edge of the gate due to the field distribution from the application of gate and drain voltages. When these voltages become large, typically > | 20 V | defects and traps in the AlGaN layer can

PAGE 36

36 occur which degrade the transport properties. This mechanism has bee n proposed and sho wn by Joh and del Alamo They proposed that once the strain and elastic energy in the AlGaN layer reaches a critical value the AlGaN layer will relax through the formation of crystallographic defects. 12,23, 50 54 Additionally, these defec ts could potentially be electrically ac tive by acting as trapping sites which explain the device degradation. 23 Degradation fro m the inverse piezoelectric effect has been corre lated with degradation of the I D and I G after a critical voltage is reached. Ad ditionally, the source and drain resistances increase and the V T shifts to a more positive value. Much research has focused on understanding the critical voltage with respect to gate length and device structure. However, it has been found that many facto rs influence the critical voltage. One of the major causes of variability in the critical voltage for similar devices, even neighboring devices within the same wafer, is non uniformity of the epilayer growth or interfacial layer between the gate and AlGaN 49,51 Another factor that affects critical voltage is the gate length. Douglas et al. reported that as gate length decreased the critical voltage decreased. 19 Addit i onally, Douglas et al. invested the effect of applied V DS on critical voltage. It was re ported that as the V DS increased, the critical voltage of the HEMT decreased. 19 Later work by Douglas et al. invest igated the effect of t emperature on critical voltage It was found that the critical voltage decreased with increasing ly applied base plate temperature to the HEMTs. 20 Another important finding on the variability of the critical voltage was reported by Marcon et al where they reported that the duration of the V GS step stress changed the critical voltage of the HEMTs. They noted that as the du ration of the V GS step stress time increased the critical voltage decreased. 55

PAGE 37

37 The degradation from the inverse piezoelectric effect has been investigated in driven onl y as a field based degradation mechanism. Their basis for this hypothesis is that the mechanical stress resultant from the applied electric field is large enough in energy to overcome the elastic energy of the crystal and relax the AlGaN epilayer causing physical degradation. 53 Due to this proposed field dependence on the degradation, current through the device should not influence this mechanism except in cases where large current density lead to device heating due to the Joule heating effect. Of course even slight heating could effect the device degradation. For example, an increase in temperature will influence the mechanical properties of the AlGaN epilayer making it more ductile and thus requiring more mechanical energy to degrade. This could potent ially slow degradation from the inverse piezoelectric effect. Of course this seems to be in contrast to work by Douglas et al. where applied base plate temperatures were associated with faster degradation. 20 dence upon the magnitude of the electric field, device and structural designs that affect the electric field profile should impact degradation. 53 For example, deposition of the gate metal or Si N x passivation can apply tensile or compress ive stress to the s urface of the HEMT thereby influencing the degradation through the inverse piezoelectric effect. These strain effects from deposition are beginning to be researched Mastro et al. simulated the strain from deposited Si N x and found it to be small in compa rison to the tensile strain already present in the AlGaN due to pseudomorphically matching the GaN. 56 Of course the deposition strain will depend upon the process conditions such as thickness, deposition

PAGE 38

38 method, temperature, pressure, and frequency. Addit ionally, the shape of the deposition itself influences the strain. For example, an opening at the edge of the gate metal results in a normal force from the gate across the surface of the AlGaN epilayer. Additionally, Mastro et al. proposed that as the ga te length of HEMTs decreases that the strain field will increase. This increased strain will need to be accounted for in future device design due to its impact on the inverse piezoelectric degradation as feature size begins to shrink for HEMTs. 2. 4.7 T hreading Dislocation Defects and Impurity Diffusion in AlGaN/GaN HEMTs Threading dislocations are almost omnipresent in AlGaN/GaN HEMTs due to their high density, ~10 9 cm 2 However, it has normally been assumed that they do not greatly affect device prop erties as AlGaN/GaN HEMTs have some of the largest electrical parameters for power and frequency applications. Unfortunately, there is growing evidence that this is no longer the case (Chapter 6 of this work will explicitly show the affect of dislocation on reliability) and that dislocations need to be taken into account for the degradation and reliability of AlGAN/GaN HEMTs Liliental Weber et al. showed that V defects or pin holes formed during the growth of GaN, InGaN, and AlGaN. 57 These defects could f orm at any depth in the growth and could thus result in a large area by the time growth was terminated due to their 56 degree opening from the geometry of their growth plans, {10 11} 57 These defects are typically much smaller with improved device quality today. However, should they form during growth they could easily make a void through the 2DEG of the HEMT which could negatively affect the device performace. Additionally, it was observed by Liliental Weber et al. that nanotubes and threading dislocat ion tend to form at the apex of these pinholes. While the reduction of pinhole size is important, the nanotubes and

PAGE 39

39 threading dislocations that are associated with them could be extremely detrimental to device performance due to providing an easier diffus ion path for impurity in the device. Pearton et al. investi g at ed the effect of dislocation density on O diffusion in GaN. An SiO 2 cap was deposited on GaN and the samples were subjected to anneals from 500C to 900C. Using secondary ion mass spectroscop y (SIMS) the diffusion profiles were compared and found to be non error function in distribution which indicated a faster diffusion than predicted by theory. This acceleration was contributed to O diffusion down dislocation cores which was found to have a n activation energy of 0.23 eV 58 This is a small value indicating that impurities could easily, rapidly diffuse down dislocations or possibly even faster down nanotube cores since they are larger in diameter than dislocations. Tapajna et al. and Kuball et al. also investigated the effect of impurity diffusion down dislocation on device degradation and reliability. Together they used trapping analysis in order to view the change in one trap amplitude against an increase in time and temperature. 18,59 It was found that as off state stressing continued the trapping amplitude increased on the order of root DT. Additionally, it was found that the trapping amplitude increased with temperature. Due to the fit of their data, they assumed a diffusion based model t o explain the increase in trapping as a result of increased impurity diffusion down dislocations. From their model, they calculated an activation energy of 0.26 eV for impurity diffusion down dislocations. 18,59 This value is slightly larger than the value from Pearton et al., but i t is still within the same order of magnitude so it could be a feasible value It is also important to note that no physical characterization was completed by Tapajna et al. and Kuball et al performed only

PAGE 40

40 minimum characterizati on to show there was no physical defect formation from stressing by TEM analysis. Thus, their calculated activation energy is based purely on a model of electrical data and no physical proof of impurity diffusion Although it is not definitively shown th at this degradation is through an impurity diffusion method it is still an important result that dislocations and impurities could potential affect device reliability. 2.4.8 Physical Degradation in Pt Gate AlGaN/GaN/HEMTs The first physical degradation no ted in the study of reliability of AlGaN/GaN HEMTs occurred in Pt gate devices. Conway et al were the first to report on physical d efect formation under the gate on HEMTs from HRL A 3000 h elevated temperature electric stress test wa s performed in air with V DS = 30 V and I DS = 40 mA. The device was heated to a simulated channel temperature of 172 C Following DC stress, the I DSS decreased 10% and G m decreased 15% from their peak values. TEM samples were prepared from the stressed devices and indicat ed that a small defect formed under the gate on the drain side of the gate and into the AlGaN epilayer of the samples. EELS analysis of the defect revealed oxidation of the AlGaN Schottky barrier layer with no evidence of gate sin king or electromigration 60 F urther w ork using the same samples along with additional non field plate T gate HEMTs w as completed by Burnham et al. Here, the T gate HEMTs were step stressed using channel temperatures of 136 C 159 C 182 C 205 C 228 C 251 C 274 C 297 C and 3 20 C through application of a base plate temperature Each channel temp erature was held for 24 h with the V DS = 15 V and V GS was adjusted to achieve an I DS = 217 mA/mm for the duration of each step stress. It is noted that HEMTs in air failed more quickly than ones in nitrogen. On average, the HEMTs exposed to air failed

PAGE 41

41 58% faster than the nitrogen ambient HEMTs. Ad ditionally, some T gate HEMTs exhibited an oxidation spike under the gate which was characterized by an O signal increase in the EDS spectru m while other HEMTs showed some possible gate sinking into the AlGaN epilaye r 61 While these results are important, it appears upon analyzing their reported TEM micrographs that most of their material is of poor quality. It can be seen that even in unstr essed samples that epilayer uniformity and roughness are poor. Therefore, these defects could have manifested due to overall low device quality causing the increase in oxidation or the roughness of the gate metal interface with the AlGaN. After an improv ement in device material quality from Triquint Semiconductor Chowdhury et al. also investigated the physical degradation of AlGaN/GaN HEMTs. Unlike previous studies, the epilayer structure was found to be smooth and flat and there was a sharp interface b etween the gate and epilayers. Here, the electrical stress conditions used were V DS = 40 V and the V GS was set so that the time zero I DS = 250 mA/mm in a nitrogen environment Additionally, base plate temperatures of 82 C 112 C and 142C were applied t o the HEMTs to raise the channel temperature to 250 C 285 C and 320C as determined by thermal simulation. These stress conditions were applied to the HEMTs for 6 h to 1000 h The unstressed device s revealed no signs of physical defect formation. Howe ver, a thi n layer of amorphous material was found between the gate and the semiconductor (this is common for feature in unstressed HEMTs ) By comparison in t he stressed HEMTs, there were many instances of physical degradation. For example, a ll stressed d evices show ed pit ting defects in the AlGaN

PAGE 42

42 epilayer under the drain side edge of the gate. These pits were typically 1 0 nm in depth which means t hey penetrated through more than half of the AlGaN epilayer. Additionally, similar pitting was also sometimes seen on the source side edge of the gate. However, the source side pitting was always significantly smaller compared to the drain side pitting. Another defect that was observed near the gate region was the formation of a crack shaped defect The crack was observed to propagate ~20 nm only from the bottom of the drain side pitting and extended through the AlGaN/GaN interface. For HEMTs with the highest applied base plate temperatures these cra cks could form in as little 6 h It is noted that rarely and in only the most extreme cases of device degradation that some gate metal was found to diffuse into the crack defects. Despite all the observation s of physical degradation from electrical stressing the most important result from this work is likely the establishment of a relationship between the percentage decrease in I DMax and the increase in physical device degradation. Here, Chowdhury et al. note d a linear dependence based on decrease I DMax and increasing relative physical degradation based on a doub le blind survey. 12 It was believed that the defects were formed by the inverse piezoelectric effect due to their positioning at the location of the largest fields in the device. Additionally, it was thought t hat combining the piezoelectric induced mecha nical strain with the tension the AlGaN layer is already under could cause the AlGaN layer to exceed the yield strain and cause this d egradation. Thus, the combination of the high temperatures from the application of the base plate, the large electric fie ld due to the applied voltages, and the initial tensile strain of the AlGaN layer could lead to the damage of the AlGaN and formation of the pit defect. Furthermore, with increasing pit depth the AlGaN becomes

PAGE 43

43 thin ner and weakened and the tensile strain could form the crack. Chowdhury et al. speculated that the crack could originate at the deepest part of the pit and spread along the width of the gate. This could explain why the crack is present in both deep and shallow p its 1 2 While this is the first paper that speculates on the origin of the physical defects in Pt gate AlGaN/GaN HEMTs, the re is no chemical analysis of the defects presented which could help explain the chemical and structural degradation. However, t he next essential work on the study o f physical degradation of AlGaN/GaN HEMTs was performed by Park et al. which included many of the same collaborators as the Chowdhury et al. work and the same device structure. Due to the similarity and the HEMTs and the collaborators the same accelerated off state DC electrical stress tests were completed as the Chowdhury et al. work. Again, here the HE MTs were subjected to base plate t emperatures of 82 C 112 C and 142C which correspond to channel te mperatures of 250 C 285 C and 320C as determined by simulation. Again, t he V DS = 40 V and the V GS was set so that the time zero I DS = 250 mA/mm However, in this is work all HEMTs underwent electrical stressing for 1000 h They noted that a lthough degraded devices have different pit sizes, the pits formed at the center of the gate finger are always larger than those at the gate finger. It was thought that temperature plays a role in the pit and defect formation as th e center of the gate finger experiences the most joule heating because it has the lo west cooling efficiency This is due to the layout geometry of the HEMT. Additionally, Park et al. completed a study comparing degradation against life time HEMTs using the electrical parameters established above were stressed for durations ranging from 6 h to 12 days. Upon completion of the s tressing, the most

PAGE 44

44 degraded 6 h HEMT was compared to the least degraded 12 day sample. It was found that the 6 h sample showed small pitting but had crack formation while the 12 day sample only had large pitting. Thus, it can be seen that the amount of physical degradation in the devices is variable a nd not consistent between HEMTs; however, it is the amount of degradation that negative ly affects the device performance and reliability. While this work again showed the formation pitting and crack defects from electrical stressing, it also presented a reaction based defect that formed under the gate. EDS analysis was applied to the defect area, and it was found that more Si, O, and C were detecte d in the defect than in the rest of the surrounding AlGaN epilayer. Also, the defect cavity that formed during degradation was backfilled with SiNx from the passivation. T he fact that the passivation filled the pit implied that increased temperature possibly due to joule he ating could be respon sible for this defect formation 15 This heating could be due to high densities of localized gate leakage leading to a very localized temperature increase a nd the defect formation. An early issue with these reliability studies on defe cts is their dependence on TEM. Due to the extremely localized area used in TEM sample analysis (<100 nm), sampling statistics and meaningful data can be difficult to gather a nd interpret. However, continuing previous work on Pt gate electrical stressing Makaram et al. used an etching sequence that exposed the top surface of the AlGaN epilayer. This allowed SEM and AFM imaging and analysis of degraded HEMTs in order to study the top structural change versus electrical stress conditions. Here, a dilute HF etch was used to removed the Si N x passivation covering the Pt gate This was followed by an 3:1 HCl:HNO 3 aqua regia etch at 80 C for 20 min to remove the source, gate, and drain

PAGE 45

45 contacts. A final surface cleaning was performed using a piranha etch sol ution consisting of 1:1 H 2 SO 4 :H 2 O 2 for 5 min at 115C. 62 This etch sequence did not damage the AlGaN surface or create artifacts that could skew the results of the SEM or AFM analysis. The only issues with the etch sequence i s that it exposed the disloc ations and pits at the top surface of the AlGaN epilayer; however, their density was low compared to the overall surface area and did not influence the results. Electrical stress conditions for the HEMTs consisted of V GS and V DG = 8 70 V at a rate of 1 V/min. Additionally, a base plate temperature of 150 C was applied to the HEMTs to simulate an increased channel temperature although the authors did not quote an estimated temperature for these conditions. For the se stress conditions, a critical voltage occurred at V DG = 20 V. Above the critical voltage, devices exhibited a permanent decrease in I DMax up to 10% and a large increase in current collapse. F or the unstressed HEMTs, the gate footprint was not visible by SEM However, for the stressed HEMTs, SEM showed a clear row of pits along the drain side gate edge. AFM analysis showed t hat in the unstressed HEMTs the top epilayer surface was smooth under the gate with no pitting defects measured. However, in st ressed HEMTs below V DG = 20 V a shallow, continuous groove formed in the drain side of the gate edge on the epilayer along the entire width of the gate Additionally, a very shallow, almost imperceptible, and discontinuous groove appears on the s ource si de of the gate As the V DG increased beyond the critical voltage, AFM showed that localized p its formed along the groove, and their density and size increased with increasing V DG The authors reported that for the most degraded HEMTs that the pits were m easured to be up to 8

PAGE 46

46 nm deep and 2 m wide. However, it is important to note that an accurate measurement of the pit dimensions is difficult with AFM. This is due to the size of the AFM probe tip being on the scale of the pit diameter or larger. Thus, t he probe size limits the resolution of the measurement making these measurements only a qualitative measurement. Makaram et al. performed an additional experiment with a constant electrical stress of V GS = 8 V and V DG = 50 V but with increasing time. Add itionally, the base plate temperature was varied. It was found that with increasing time and/or temperature that the pit density and size increased and the current collapse and reduction in I DMax increased Due to the dependence of the degradation on tem perature and field magnitude, It was postulated that the pitting grooves and pits were formed by a field induced diffu sion process or an electrochemical etching of the epilayers. Continuing the study of under gate pit formation, Li et al. used a high power on state stress on Pt gate HEMTs. Here the V DS = 40 V and the V GS was adjusted so that the I D remained at 250 mA/mm for 523 h which r esulted in a 33% in I DMax 63 Additionally, a base plate temperature of 120 C was applied to the HEMTs. Using these condi tions, thermal simulations estimated the channel temperature to be ~354 C Similar to the off state stress, pits formed along the drain side of the gate. It is noted that the pits merged together more frequently at the center of the device than at the en d with emerged pits forming a trench 50 nm wide and 8 nm deep. Again, the previously noted limits of AFM resolutio n apply to these measurements. It was proposed by Li et al. that the increased merger in pitting at the center of the HEMT finger was due t o the increased temperature and stress in the middle of the

PAGE 47

47 finger compare d to the end. Additionally, it is noted that pitting was greater under fingers located in the middle of the device compared to the outside of the device This is cons istent with a higher temperature effect due to reduced heat dissipation from the geometry of the device layout. The geometry of the device layout and bias conditions were used to create a thermal simulation of the temperature distribution of each finger on the device. The simulated temperatures of the fingers were compared to the pit area to the 3/2 power Using an Arrhenius extrapolation, a calculated activation energy of 0.32 eV for the pitting process was found. This value is similar to previous diffusion process results on GaN. The re are two important results from this experiment. F irst t hat during on state stress pitting can occur at operating temperatures Second that this pitting is non uniform based on the device geometry which could suggest that it is ac celerated by the self heating of the device during operation. 2.4.9 Physical Degradation in Ni Gate AlGaN/GaN/HEMTs As previously detailed in the above section, there is a lot of work reported in the literature on physical degradation of Pt gate AlGaN/GaN HEMTs. However, there is comparatively little work reported for Ni gate HEMTs. Due to the almost singular use of Ni gate HEMTs in this work, one of the biggest impacts of the research presented in subsequent chapters will be the additional understanding of physical degradation of Ni gate AlGaN/GaN HEMTs added to the body of knowledge of the field. However, first a review of reported physical degradation of Ni gate AlGaN/GaN HEMTs will be summarized. The presence of O in the ambient during electrical str essing has been shown to decrease Id and increase current collapse of AlGaN/GaN HEMTs quicker compared to higher vacuum conditions 64 Gao et al attributed this degradation to the formation of

PAGE 48

48 defects along the edge of the ga te 64 In their work, AlGaN/GaN H EMTs passivated with Al 2 O 3 were electrically stressed using off state conditions of V DS = 0 V and V GS = 40 V for 60 s, 600 s, and 6000 s. As the duration of the stress increased, the amount of e r stressing, the gate metal and stringers were etched and revealed pits ranging from 25 to 45 nm in depth around 2 O 3 passivation and the Ni gate. It t ambient O due to the large electric field. When the se stress conditions were repeated in a vacuum condition of 3 x 10 5 Torr, the pit f ormation was significan tly reduced. Additionally, these devices showed 40% less permanent degradation and the current collapse was 30% less than in HEMTs stressed in ambient atmosphere. The decrease in degradation was attributed to the lower density, siz e, and depth of the pits compared to stressed in ambient samples. However, in both stress conditions the V T shifted positively by 4 V. Gao et al hypothesized that the large electric field at the gate edge accelerated this pitting by lowering the activati on barrier for oxidation To test this, two different off state stress conditions were used. In both conditions the V GS = 12 V but in one the V D = 30 V and V S = 0 V while in the other V D = 0 V and V S = 30 V. By reducing the Vg and increasing the V D or V S a non symmetric electric field is produced around the gate. This is dissimilar to the previous stress state with V D = 0 V and V G = 40 V where the field was symmetric and pitting occurred on both sides of the gate. These two stress

PAGE 49

49 states showed that side with the 30 V indicating that they only formed on the high field side of the gate. Work by Whiting et al also used removal of the gate metal via etching to view under gate pitting in Ni gate HEMTs using off state electrical stressing of V G = 8 V and V DS = 0 70 V in 1 V/min steps. 21 During the stressing, a jump in the Ig occurred. It was found after gate metal removal and top down SEM that pitting defects occurred under the mid dle of the Ni g ate only after the Ig jump. This stress condition was used for both 100 nm a n d 1 m HEMTs. At the end of stressing, the final Ig was recorded and compared to the fractional gate area of the HEMT that showed pitting. It was found that an increased area of under gate pitting correlated with a high final I G Additionally, for the same final I G the 100 nm HEMTs showed more pitting compared to 1 m HEMTs. Unfortunately with these pitting studies, TEM of the pitting prior to etching was nev er obtained. Thus, no compositional information on the pit was obtained and only speculation on chemistry due to oxidation or gate metal diffusion was possible.

PAGE 50

50 Figure 2 1. Schematic of wurtzite GaN A) Ga face. B ) N face. Figure 2 2. XTEM mic rograph showing the typical structure of an AlGaN/GaN HEMT.

PAGE 51

51 CHAPTER 3 EXPERIMENTAL TECHNIQUES The sections of this chapter serve as a basic introduction to the essential and relevant material processing and characterization techniques that were used t o complete this work. Advanced theories and de tails are omitted, but a cursory review of the techniques is p resented for familiarity. If processing procedures are consistent for a technique then they are presented in this section; however, for processes with variability specific methodologies will be presented for each technique in the experimental section of the following chapters. 3.1 Material Processing 3.1.1 Thermal Annealing In order to isolate only th e effect of thermal stress on AlGaN/GaN HEMTs, transistors were placed in a standard tube furnace to undergo thermal anneal. All anneals used temperature was regul ated by a heating coil wrapped around its perimeter as shown in Figure 3 1 Due to the miniature size of th e reticule that containe d the transistors (~ 7mm by 3mm), transistors were placed on a cleaved piece of Si (100) wafer that was wedged into a slot on a preheated quartz boat. This setup prevented the transistors from moving during the insertion and extract ion of the boat from the furnace and allowed accurate measurement of the temperature. The boat was always placed in the sa me exact spot for each anneal near the center of the furnace using a system of rods and markings. Because of the size of the samples the temperature was not measured directly on the sample. Instead a thermocouple was positioned on the Si (100) wafer holding the transistors without touching the transistors the quart z boat, or any part of the quartz tube. Furthermore, the temperature was

PAGE 52

52 measured both at the start and at the finish of each anneal. Additionally, t he thermocouple measurement error is estimated to be 1C as per the accuracy of th e device. It is noted that one end of quartz tube had a gas inlet and the other end was cap ped to permit the application of different ambients. For this work, the anneal temperature was varied from 100 C 850 C a nd for durations from 0.25 h 1000 h. Generally, this method is considered to be slow for sample temperature ramp up and cool down (< 5 min) compared to a rapid thermal anneal. However, since the ramp up and cool down time was much smaller than the duration of the anneals, this possible error is minimized and the measured anneal times are taken to be accurate. 3.1.2 Transistor Etchi ng In order to observe any large areas (dimensions greater than 20 nm x 20 nm) of physical degradation on top of the AlGaN surface, the passivation and gate metal of the HEMTs were removed by wet chemical etchin g The first step of the etch process is the removal of the Si N x passivation layer covering the AlGaN/GaN HEMTs by submerging them in a buffered oxide etch solution of HF for 15 min. Next, the HEMTs were clean ed with water in order to remove any residual solution. After the cleaning step, the meta llization was removed by submerging the HEMTs in a solution of FeCN and KI for 0.75 5 days. Following the removal of the metal, the top surface of the AlGaN epilayer was exposed to ambient. This surface was cleaned of any build up of reaction products from the etching steps by ultrasonication in solution of acetone and n heptane for 1 h, then methanol for 1 h, and lastly in water for 1 h. It has previously been shown that epilayer of the HEMTs for periods up to at least 18 h. 21

PAGE 53

53 3.1.3 XTEM Sample Preparation TEM requires electron transparent samples. However in order to achieve electron transparency, TEM foils typically must be < 300 nm thick and preferably < 100 nm thick if any chemical characterization is desired or if the sample is composed of high Z elements. A thinner foil is advantageous because as thickness increases the number of scattering events (both elastic and inelastic) increases making analysis more difficu lt. TEM foil preparation typically involves using either an ion mill or a focused ion beam (FIB) system. In this work, a FIB system was used to prepare site specific cross sectional TEM (XTEM) foils due to the required site specificity involving regions of interest (ROI) < 100 nm wide. This type of accuracy is not possible in an ion mill which typically requires a large, uniform sample such as a wafer. Although, ion mills use lower voltages and currents than FIB systems resulting in foils that are less damaged and therefore capable of higher resolution imaging, they take significantly more time to prepare a FIB foil. In this work, the FIB system used for XTEM sample preparation is a dual beam machine with both an SEM and an ion beam which are off set 5 2 from each other. This orientation permits SEM micrographs to be gathered during the milling progress without damaging the sample. For FIB preparation, the HEMTs were placed on Al stubs using C adhesive tape. If the HEMTs were previously etched (Secti on 3.1.2), then ~100 nm of C was thermally evaporated using a carbon rod evaporator system onto the surface to make the surface conductive for imaging and to protect the surface during subsequent stages of FIB preparation. If the HEMTs were not previously etched, then no C evaporation was necessary for sample conductivity and surface protection due to

PAGE 54

54 the thick Si N x passivation layer covering the transistor. Once the samples are placed in the FIB system and the vacuum reaches < 7 x 10 5 mBar, then a focus ed beam of Ga + ions is accelerated at 30 keV towards an ROI. A gas injection system is used to deposit an organo Pt layer ~ 8 m wide x 2 m long and 1.8 m thick over the ROI using a 300 pA beam current. This Pt layer protects the ROI from further ion b eam damage as the sample is milled thinner later in the process. After Pt deposition, large, deep trenches ~ 10 m wide x 7 m long x 3.5 m deep were milled above and below the Pt using a 5000 pA beam current which exposed the cross section of the HEMT a s shown in Figure 3 2 A B. The Pt covered ROI is then progressively thinned using a 1000 pA beam current mill until its thickness reaches ~1m causing the sample to resemble a wish bone shape. This is followed by an undercut ~ 2m below the surface o f the ROI which separates the ROI from the substrate using a 1000 pA beam current. Then another 1000 pA beam current mill is used to release one side of the sample from substrate as shown in Figure 3 2 C. A micromanipulator actuating a W needle is brou ght close to the surface of this free side and is attached using organo Pt deposition as shown in Figure 3 2 D. Once the sample is attached to the W needle, the other end is mil led to l release the sample from the substrate as shown in Figure 3 2 E. With the sample attached to the W needle, it is vertically moved away from the surface of the HEMT and moved to a Cu grid with fingers. The W needle positions the sample so that it touches the grid and the W at the same time, but without overlapping the C u grid and the ROI as shown in Figure 3 2 F. Then the sample is adhered to the Cu grid using the organo Pt deposition and is milled free from the W needle as shown in Figure 3 ng a

PAGE 55

55 50 pA beam current until the thickness < 300 nm as shown in Figure 3 2 H. Then the sample is tilted 2 on each side and milled which reduces the thickness < 100 nm. For the final thinning step, the sample is rotated 3 and milled so that it forms a wedge with the ROI positioned slightly before the edge of the sample achieving a thickness < 100 nm as shown from a top view of Figure 3 2 I. At this point, the XTEM foil is complete is ready for imaging and analysis. 3.1.4 APT Sample Preparation Sim ilar to TEM sample preparation APT sample preparation requires the fabrication of samples on the nm scale The samples are fabricated either by mechanical electropolishing or a FIB method to form sharp needle like tips ~100 nm in diameter, 100 nm above t he ROI, and possessing a half shank angle between 5 7 In this work, two different APT sample preparation techniques were used where the ROI was either mounted on a Si post array or a W wire. These techniques will be covered in the following sections 3.1.4.1 Si post array mou n ting technique The steps to mount a ROI onto a Si post are the same as the steps for XTEM sample preparation in Section 3.1.3 until the sample is extracted from the substrate attached to the W needle as shown in Figure 3 2 E. Following sample extraction, Figure 3 3 A B shows the ROI is centered on top of a Si post where it is bonded to the post using organo Pt in Figure 3 3 C. Then 2 m widh sample of the ROI is cut free from the rest of the ROI and the backside of the ROI is bonded to the Si post again using organo Pt as shown in Figure 3 3 D. The Si post is then tilted to 52 so that the ROI is parallel to the ion beam as shown from top view in Figure 3 3 E and from 52 in Figure 3 3 F. An annular mill at 5000 pA with an inner radius of 4 m and an outer

PAGE 56

56 radius of 8 m centered around the ROI is used to mill the Si pillar as shown in Figure 3 3 G. A second annular mill at 1000 pA with an inner radius of 1 m and an outer radius of 4 m centered around the ROI is used to mill the Si pillar as shown in Figure 3 3 H. This is followed by a third annular mill at 300 pA with an inner radius of 0.3 m and an outer radius of 4 m centered around the ROI to mill the Si pillar. This mill is continued until the tip a pex is ~500 nm from the ROI. By this time the Si post and ROI should be needle like in shape. Finally, the FIB accelerating voltage is reduced to 8 kV and a continuous low voltage mill is used to clean up the tip surface from damage caused by higher volt age milling. This mill is continued until the tip apex is ~100 300 nm from the ROI as shown in Figure 3 3 I. 3.1.4.2 W wire mounting technique An electropolishing technique was used to make W needle substrates on which the ROI was mounted for APT an alysis. To make the W needle carrier substrates, W rods were cut and attached to the positive terminal of a power source. The negative terminal was connected to a metal hoop which was placed into an electrolyte solution. The W rod was then cyclically di pped into the solution and removed in 3 s intervals until the tip appeared needle like in shape. A schematic of the system used to electropolish t he W wire is shown in Figure 3 4 Although electropolishing was used to make W needle carrier substrates, FIB methods were used to attach and fabricate the ROI into a tip for APT which were similar to the instruction in the previous section 3.1.4.1 3.2 Material Characterization 3.2.1 Electrical All electrical measurements and stres sing for the AlGaN/GaN HEM Ts were completed using a HP 4156C semiconductor parameter analyzer except for C V

PAGE 57

57 measurements which were measured using a n Agilent E4980A precision LCR meter The HEMTs were unpackaged and electrical measurements and applied electrical stresses were com p leted using probe tips. Conducting metal needles were brought into contact with the contact pads of the devices. Then voltages were applied to the needles in order to generate electric fields and current flow. Figure 3 5 shows the setup for the TLM m easurement that was used to extract the contact resistance from the HEMTs. When the resistance measured between the pads is plotted against the pad separation, the contact resistance can be found by the y intercept squared and multiplied by the length of the contact pad, but it is divided by the product of the slope of the plot by 4 by the width of the contact pad. 3.2.2 SEM SEM is a quick method for obtaining high magnification images of relatively large areas of a sample as conductive samples can be imme diately imaged while nonconductive samples merely need a thin layer of conductive coating deposited over their surface. Compared to optical microscopes SEMs have superior depth of field and higher resolution This is due to the wavelength of the electro ns used for imaging which are smaller than the wavelengths of visible light. The electrons are either thermionically or field emitted from the gun and accelerated towards the ROI between 1 40 k e V. The majority of the SEM images taken in this work were at 5 kV. While traveling towards the ROI the electrons are focused by a series of magnetic lens. Additionally, the last set of magnetic coils manipulates the beam in the x or y direction on the ROI surface called rastering which produces the SEM imag e signal. The signal is created by the detection of secondary (SE) or backscattered electrons (BSE) emitted from the interaction volume created from the electron beam

PAGE 58

58 penetrating into the ROI. In SEM the SE and BSE are measured by different detectors and due to the nature of their origin reveal different information about the ROI. SE are generated near the top of the intera ction volume due to inelastic collisions between the electrons and near surface atoms in the ROI. These collisions cause the emis sion of s These emitted electrons are then collected by a detector which creates contrast based upon the number of SE detected. Thus, more emitted SE correspond to brighter features in the ROI. A dditionally, as the angle between the electron beam and surface become s shallower or as the beam is brought closer to an edge feature in the ROI, the number of SE emitted will increase due to their c loser proximity to the surface. This causes edge feature s in the ROI to appear brighter which makes detection of SE a good method for observing the topology of the ROI. A second type of contrast mechanism in SEM involves BSE. BSE are electrons from the incident electron beam that are elastically scattered by the atoms in the ROI. With increasing atomic number (Z) of the atoms in the ROI, the larger the backscattering cross section becomes. Thus, higher Z atoms produce more BSE and appear brighter in contrast when using a BSE detector compared to lower Z ele ments. This permits comparis on of compositional gradients or grains in a ROI. 3.2.3 TEM Like SEM, TEM uses accelerated electron s to image a ROI; however, unlike SEM the sample preparation is time intensive, the ROI is on the m scale, and the sample thic kness is < 100 nm This small sample size makes TEM analysis prone to statistical aberrations and requires complex analysis if the bulk material analyzed is inhomogeneous. However, the resolution and analytical capabilities of the TEM are

PAGE 59

59 unparalleled by any other microscope. Additionally, the small sample size and high resolution makes TEM the optimal and most frequently utilized tool for the analysis of the HEMTs p resent in this work. For the TEM in this work, the accel erating voltage was set to 200 k V and the electron beam was generated using field emission. It is noted that thermionic emission is possible for the electron source in TEM Additionally, typical operating voltages normally range from 100 300 kV. Like SEM, using an electron beam per mits the resolution below the wavelength of visible light; however, in TEM the accelerating voltages are one to two orders of magnitude larger than in SEM which help s TEM have a much higher resolution. The derivation of the wavelength of an electron is po ssible through the wave particle duality of matter first proposed by Louis de Broglie where an electro n with momentum, p e can be considered wave like in nature possessing a wavelength, where: (3 1 ) where h = 6.626 10 34 The kinetic energy of an electron can be calculated as follows: E = m e v 2 (3 2) where m e = 9.109 10 31 kg is the mass of an electron and v is the velocity of the electron. Additionally, the momentum of an electron is know n to have the following relationship with its mass and velocity: p e = m e v (3 3)

PAGE 60

60 therefore it can be shown that the kinetic energy and electron momentum are relat ed through: E = p e 2 m e 1 (3 4) By u sing the conversation of energy, it can be shown that kinetic energy of an electron is equal to the energy an electron gains passing through a potential difference (applied voltage) establishing relation ship betwee n p e and V e which is given by: (3 5 ) where e = 1.602 10 19 C is the electric charge of an electron. Com bining Equations (3 1 ) and (3 5 ), it is easily shown that : (3 6 ) Therefore, if an electron were accelerated at 200 k e V it would have a ~ 2.7 10 12 m or 2.7 pm if relativistic effects are neglected. However, above 100 kV relativistic effects need to be taken in to account which reduced the to ~ 2.5 pm instead of the original value of 2.7 pm. After the electron is generated from the source, it passes through the condenser lens and aperture. The condenser lens controls the minimum spot size of the beam and focu ses the beam on the sample. Additionally, the condenser aperture controls how much of the beam goes to the sample and therefore controls the intensity of the beam. The electron beam then passes through the sample where it undergoes both elastic and inela stic scattering. After passing through the sample, the beam is focused by an intermediate lens and then projected by the projector lens system onto either a

PAGE 61

61 phosphor screen or a CCD camera. A schematic of the internal set up for a TEM is shown in Figure 3 3. Depending on the analysis it is possible to select either the transmitted or diffracted parts of the beam using an objective aperture. By selecting specific diffracted beams, increased contrast along specific crystallographic orientations is realize d The TEM micrographs produced in this work acquired in both bright field and dark field mode using both conventional TEM and scanning TEM (STEM). 3.2.4 APT Atom probe tomography (APT) is a characterization technique where ultra sharp tips of materials ~ 100 nm in diameter are fabricated and field evaporated. The atoms are collected by a tiered, multi channel detector. This allows a time of flight (TOF) detection of mass to charge (M/C) ratio for the field evaporated ions. APT is a progression from fie ld ion microscopy (FIM) which is a point projection technique where an imaging gas is used to make an image of the surface Typically in FIM, a tip is cooled below 78 K an d an imaging gas such as He is used in the chamber. 65 An electric field is generate d at the tip and can be calculated by: (3 7) where F is the electric field, V is the voltage applied, R is the tip radius, and is a geometric factor. 66 Typically, the field is on the order of 15 60 V nm 1 when the tip is biased. T hen t he He ab sorb s onto the tip surface and loses an electron due to tunneling via a quantum effect. 67 The now positively charged He ions are electrostatically r epelled orthogonally from the t ip surface and detected by the multi channel plate. 65 This produces a recordable image of the surface.

PAGE 62

62 The transition from FIM to APT occurred when a large DC bias was applied to the tip. With this increasing bias, the electric field continued to rise. Eventually, the electric field became large enough and atoms on the tip surface began to sublimate from the surface removing the dependence of an imaging gas T his field evaporation of atoms due to a large electric field is a material property that varies with each element and compound. 6 8 Thus, different compositions requ ired different voltages in order to achieve field evaporation. This i s an important concept because if the voltage is too large, the field evaporation will be too rapid causing the tip to fracture However, using a voltage too small results in no field e vaporation Furthermore, the modern APT development occurred with the addition of the delay multi channel TOF detector 69 Due to this design, the TOF of ions could now be mea sured and therefore the depth from the tip surface to the detector could be measur ed. This allowed APT to become a 3 D analysis technique compared to FIM which is a 2D surface limited technique. Additionally, the TOF detector permitted the chemical analysis of each evaporated ion due to the M/C measurement. Combining the 3D aspect of the analysis with the capabilities of the chemical analysis has allowed the creation of 3D chemical map reconstruction s, which has permitted APT to become a powerful too l for material characterization 70 72 Furthermore, APT has a theoretical resolution tha t rivals TEM. For example, the best achievable spatial resolution is ~0.02 nm for metals. Of course this resolution is materials dependent based upon the evaporation quality of the sample (more details will be provided about this later). However, even i n Si, the (200) planes have been resolved using APT which would result in a resolution of ~0.27 nm. 65, 73 74

PAGE 63

63 The field evaporation of the tip depen ds on its constituent material. Originally, APT was limited to metals due to their high electrical and therma l conductivity. These properties allowed metals to evaporate easily and not fracture easily due to the large electric fields. However, the evaporation of semiconductors an d insulators remained elusive. The E G of the materials created a larger barrier to field evaporation that was difficult to overcome using only DC or pulsed biasing. Additionally, semiconductors and insulators typically have poorer electrical and thermal conductivity compared to metals which lowers the evaporation quality even when fiel d evaporation occurs. The realization of the field evaporation of non metals including semiconductors, insulators, and polymers occurred with the additional of a pulsed laser applied to the tip. The laser pulse imparts additional energy to the tip with p hotons allowing it to over come the field evaporation barrier 66,73,75 Finally the application of different laser wavelengths has improved the efficiency of the field evaporation of non metals. Today ROI of everything electrical and thermal conductivity, E G and size can be field evaporated. Recently, thick oxide films, polycrystalline Si, gate oxides, dopant interactions and clustering in semiconductors silicide contacts, AlGaN/GaN HEMTs, and FinFETs have been successfully field evaporated using laser p ulsing. 76 92 Also it is noted that t ypically, the larger the E G of the materials the smaller the applied laser wavelength needs to be in order to achieve efficient field emission of the sample.

PAGE 64

64 Figure 3 1. Schematic of the tube furnace set up us ed for thermal annealing

PAGE 65

65 Figure 3 2. SEM and ion beam micrographs showing the FIB procedures to lift out a XTEM sample of an AlGaN/GaN HEMT. A) Ion beam image of Pt deposition and one milled trench. B) Ion beam image after completion of both mille d trenches. C) Ion beam image showing the freeing of one side of the sample from the bulk substrate. D) Ion beam image showing the attachment of the micromanipulator to the sample. E) Ion beam image of the sample extracted from the substrate. F) SEM im age of the sample attachment to the Cu grid G) SEM image from 52 of the sample attached to the Cu grid H) SEM image showing the progression of the sample thinning I) Ion beam image from top view showing the sample thickness after final thinning

PAGE 66

66 Figure 3 3. SEM and ion beam micrographs showing the FIB procedures to fabricate an APT tip of an AlGaN/GaN HEMT. A) SEM image from top view showing the sample alignment over the top of the Si post B) Ion beam image from 52 showing the sample alignme nt over the Si post C) Ion beam image showing the attachment of the sample to the Si post D) Ion beam image showing the detachment of the sample from the micromanipulator E) Ion beam image from top view of the mounted sample on the Si post F) SEM i mage from 52 showing the sample shape before annular milling G) SEM image showing the tip shape after first annular mill H) SEM image showing the tip shape after second annular mill I) SEM image showing the tip shape after final mill

PAGE 67

67 Figure 3 4 Schematic of the electropolishing system used to sharpen the W wires into sharp tips.

PAGE 68

68 Figure 3 5. Diagram of the setup of the TLM measurement technique. Here, x i is the spacing between the contact pads, Z is the width of the contacts, and L is the length of the contacts.

PAGE 69

69 Figure 3 6 Schematic of the internal configuration of a TEM.

PAGE 70

70 CHAPTER 4 CHARACTERIZATION OF GATE INTERFACIAL LAYERS 4.1 Effect of Interfacial Layers on HEMTs R eliability and degradation issues remain a significant cha llenge and concern for future device improvement in AlGaN/GaN HEMTs 12 21 In order to accurately understand how defects change the internal structure of the device and degrade its properties, the initial state must be known in as much detail as possible. Due to a continually maturing process development technology, AlGaN/GaN HEMTs always have some type of native defect present such as dislocations, pinholes, or interfacial layers between contacts. S pecifically, it is known that interfacial layers between contacts and epi layers can have considerable effects on device performance. One particular area of concern is the interface between the gate metal stack and AlGaN epi layer. It has been shown that variation in th e interfacial layer thickness and chemistry cause the SBH to change. 47 Additionally, decreasing the thickness of this interfacial layer from annealing was found to improve the device reliability during life time testing. Therefore, accurate chemical chara cterization of the gate metal stack/AlGaN interfacial layer, which may have nm to sub nm thickness, may prove crucial in understanding and ultimately improving HEMT reliability and performance. By knowing the original composition of the interfacial layer and how it changes due to processing or stress, the change in electrical properties can be measured. This would allow for a structure property relationship to be established. Here, a combin ation of HAADF STEM and APT were used to characterize a Ni/AlGaN i nterfacial oxide layer Parts reprinted with permission from M.R. Holzworth, N.G. Rudawski, S.J. Pearton, K.S. Jones, L. Lu, T.S. Kang, F. Re n, and J.W. Johnson, Characterization of the gate oxide of an AlGaN/GaN high electron mobility transistor, Applied Physics Letters. 98, 122103 (2011), Copyright 2011, American Institute of Physics.

PAGE 71

71 with sub nm thickness. Additionally, it was found that the interface thickness influenced the field evaporation of the HEMT APT tips. Thus, multiple device geometries and APT evaporation conditions were used in order to analyze in terfacial layers 4.2 Experimental In this work three types of AlGaN/GaN HEMT growth structures were used with varying gate lengths ranging from 100 nm to 1 m. The three HEMT structures had different thicknesses for their interfacial layer between the ga te/AlGaN epilayer Additionally, two types of APT systems were used to analyze the structures. More details on the device structures and APT systems are presented in the following sections. 4.2.1 Device Structures 4.2.1.1 Si(111) substrate Ni gate HEMT The semiconducting epi layers in the AlGaN/GaN HEMTs used for this work were grown on 100 mm Si (111) substrates using MOCVD with a scheme described in more detail elsewhere 3, 47 93 T he nominal composition of the AlG aN was Al 0.26 Ga 0.74 N. The fabrication steps following epi layer growth to create the HEMT structures (including Ni/Au gate metal stack Schottky contact formation) are described in detail in Ref erence 2. 4.2.1.2 SiC substrate Ni gate HEMT For the AlGaN/Ga N HEMTs grown on SiC used for this work, all were grown on the same semi insulating 6H SiC substrate and received the same processing. An AlN nucleation layer was used on the SiC and followed by a 2.25 m thick Fe doped GaN buffer, 15 nm of Al 0.28 Ga 0.72 N, and capped with 3 nm of an unintentionally doped GaN layer (this layer is not normally visible in TEM micrographs). The ohmic contacts

PAGE 72

72 consisted of metal stacks of Ti/Al/Ni/Au annealed for 30 s at 850 C. 1 6,1 9 20 Additionally, 150 m width and a gate length of ~100 nm ~170 nm, or ~1 m. 4.2.1.3 Pt gate HEMT With the Pt gate HEMTs, devices were grown on SiC using MOCVD. This was followed by an epitaxial GaN buffer layer. On top of the buffer, a 16 nm thick Al 0.28 Ga 0.72 N layer was grown and topped with a 3 nm GaN cap. The Pt gate employed a double layer Pt/Au design with the Pt in con tact with the epilayers. Finally the device was passivated with Si N x 4 .2.2 APT Systems Two types of APT systems were used in this work. The system experimental parameters used for each system are described in the following sections. 4.2.2.1 Image LEAP tomograp hy system Laser assisted APT was performed using an Imago local electrode atom probe (LEAP) 3000X Si system. During field evaporation, the specimen temperature was maintained at 65 K with a chamber pressure < 3.2 10 9 Pa, and the laser wavelength was 532 nm with a pulsing frequency of 250 kHz. 4.2.2.1 Cameca APT system For APT analysis on the Cameca system, an Amplitude Systems s pulse femtosecond Yb:KGd(WO 4 ) 2 laser was used to laser assist the field evaporation. A locally built straight type 3DAP instrum ent was fitted with a CAMECA fast delay line detector. The flight path of the ions was 130 mm and the acceptance angle of the CAMECA delay line detector was ~ 0.3 sr. For the experiments, t he temperature

PAGE 73

73 ranged from 22 K to 60K, the l aser energy was varie d between 0.4mW to 10mW, and the ambient consisted of ultra high vacuum conditions with the occasional addition of a s light pressure of Ne The slight addition of Ne was added to help reduce the electric field at the sample tip. Additionally, while keepi ng the temperature, laser power, and ambient at the same conditions, the orientation of the gate region with respect to the tip apex was changed in order to determine the effect of sample orientation on the quality of the field evaporation. 4.3 Results and Discussion 4 .3 .1 Structural TEM Analysis XTEM samples of the HEMT structures were prepared using FIB milling methods described in Section 3.1.3. Here, l ow magnification HAADF STEM images of the gate stack and semiconductor for the Ni gate on Si(111), Ni gate on SiC, and Pt gate are presented in Figure 4 1 A C respectively. I n these images, brighter features corresponding to areas of greater average atomic number; the individual Au, Ni, AlGaN, and GaN layers are indicated along with the approximate a rea used for high magnification HAADF STEM images shown in Figure 4 1 D F and APT analysis shown in Figure 4 2 A C. Comparing the high magnification HAADF STEM images from Figure 4 1 D F, it is clear that the interfacial layer (dark band) betw een the gate metal and AlGaN epilayer increases as the samples progress from Ni gate on Si(111) to Ni gate on S iC to Pt gate on SiC. Here the interfacial layer thickness increases from 5 and finally to 18 respectively for the samples. 4.3.2 Imago APT Analysis APT samples of the HEMT structures were prepared using FIB milling methods described in Secti on 3.1.4. APT reconstructions of the gate/AlGAN interface are shown

PAGE 74

74 in Figure 4 2 A C for the Ni gate on Si(111), Ni gate on SiC, and Pt gate, respectively. Figure 4 2 A presents a reconstruction of the collected data from a single APT sample from the Ni gate on Si(111) device showing the location of different individual atoms. The data reveals that the Ni/Au gate metal stack and AlGaN/GaN epilayers were field evaporated in sequential order, therefore implying inclusion of the Ni/AlGaN interfacial laye r in the analysis Due to the quality of the field evaporation from the Ni gate on Si(111) samples, the complete M/C and one dimensional ( 1D ) concentration profiles were acquired. Although the reconstruction of the Ni gate on Si(111) was complete, it ca n be seen in Figure 4 2 B and C that the reconstructions of the Ni gate on SiC and Pt gate are incomplete and the entire gate/AlGaN interface did not successfully field evaporate. To provide more evidence of the incomplete field evaporation, the evaporati on mass history from the three reconstructions in Figure 4 2 A C are shown in Figure 4 3 A C, respectively. From Figure 4 3 A which depicts the Ni gate on Si(111) which was successfully field evaporated, it is clear that the mass histories of the dete cted ion species are continuous, broad, and graded. This indicates that the layers are being evaporated because of the data transitions between the different layers. This is in contrast to Figure 4 3 B and C where the mass history is very sharp and disco ntinuous at the interface between the gate and AlGaN epilayer. This indicates that a fracture occurred in the gate metal, but field evaporation was restarted somewhere in the AlGaN and GaN layers. Due to this incomplete field evaporation, all the layers were not detected in Figure 4 2 B and C. Specifically, the bottom of the Ni layer of the gate, the interfacial layer between the gate and AlGaN, and the top of the AlGaN epilayer were

PAGE 75

75 not detected in Figure 4 2 B and C. Thus, complete M/C spectra and 1D concentration profiles for the Ni gate on SiC and Pt gate on SiC are not presented because they could not be acquired. However, due to the high quality of the field evaporation from the Ni gate on Si(111) samples, the complete M/C and 1D concentration prof iles acquired were analyzed. Here, the complete M/C spectrum (not presented) of the region indicated in Figure 4 1 A indicates peaks at 21.5 and 74.0 that correspond to 27 AlO ++ and 58 NiO + respectively, which are presumably constituents of the interfacial oxide layer. A partial M/C spectrum shown in Figure 4 4 A shows three peaks at 41.5, 42.0, and 42.5; these peaks correspond to 69 GaN ++ 14 N 3 + and 71 GaN ++ or 69 GaO ++ respectively. It is important to note there is no distinguishable peak at 43.5, which corresponds to 71 GaO ++ Additionally, the partial M/C spectrum presented in Figure 4 4 B shows no distinctive peaks at 85.0 and 87.0, which correspond to 69 GaO + and 71 GaO + respectively. The lack of a peak at 85.0 and the presence of a peak at 42.5 indica te the possibility that the peak at 42.5 may be due to 71 GaN ++ instead of a 69 GaO ++ Distinguishing between 71 GaN ++ and 69 GaO ++ can be accomplished by evaluating the isotopic ratio of 69 Ga/ 71 Ga. If there is no significant amount of 69 GaO ++ at the 42.5 pe ak, then the ratio of the value between the peaks at 41.5 and 42.5 should correspond to ratio between 69 GaN ++ and 71 GaN ++ which is correlated to the isotopic ratio between 69 Ga and 71 Ga, ~1.507. Therefore, if a significant amount of 69 GaO ++ is present, th en the ratio between the peaks at 41.5 and 42.5 should be lower than the natural isotopic ratio.

PAGE 76

76 The peaks at 41.5 and 42.5 are slightly non Gaussian in shape; instead of possessing a true, center maximum, each peak has a slight plateau where there are mul tiple maximums that may or may not be centered. To compensate for this shape, the average value over the plateau was calculated and used as the number of counts at the 41.5 and 42.5 peaks. The average ratio between the peaks at 41.5 and 42.5 was calculat ed to be 1.516 0.05; this value is within measurement error of the natural isotopic ratio of 69 Ga to 71 Ga. With the measured ratio very close to the natural isotopic ratio, it is reasonably concluded that 42.5 is most likely a 71 GaN ++ peak instead of a 69 GaO ++ peak. Thus, with the peak at 42.5 associated with 71 GaN ++ there appears to be no GaO x detected in the HEMT. 1D concentration profiles of Au, Ni, O, Ga, N, and Al across the gate region into the semiconducting epilayers were obtained from the full reconstruction presented in Figure 4 2 A using a 40 nm cylindrical data pipe. The data pipe was positioned orthogonally to the Ni/AlGaN interface by using the iso concentration curve of Al. By positioning the data pipe orthogonal to the interface, the r esulting 1D concentration profile is more accurate because the smearing between the layers is minimized. The complete 1D atomic concentration profile for the gate region of the Ni gate on Si(111) HEMT is shown in Figure 4 5 A indicating an O peak at the N i/AlGaN interface, confirming the evaporated O + detected in the full M/C spectrum resides in the Ni/AlGaN interfacial layer. Peaks on the full M/C spectrum at 21.5 and 74.0 were previously identified as 27 AlO ++ and 58 NiO + ; the 1D concentration profiles of these ions are presented in Figure 4 5 B in addition to the measured O + profile in the interfacial layer. The peaks of the 27 AlO ++ and 58 NiO + ions are distinct and do not overlap within the O +

PAGE 77

77 profile indicating the interfacial layer is composed of disti nct AlO x and NiO x layers adjacent to the AlGaN epi layer and Ni gate contact, respectively. The observation that the interfacial oxide is composed of distinct NiO x and AlO x layers is an important result and provides insight into possible device failure mec hanisms. Specifically, it is known that AlGaN oxidation can occur even under ultra high vacuum conditions 94 and the oxidized layer consists of non stoichiometric O rich AlO x 95 98 The excess O from the AlO x layer can therefore be used to form an adjacent NiO x layer during or a fter Ni deposition, which changes the composition of the AlO x layer. The generation of the NiO x layer and/or the compositional changes to the AlO x layer after NiO x generation may change the electrical properties of the gate/channel interface, particularly the surface Fermi level, SBH, and surface band bending, which will influence device performance and may influence device failure. While this proposed formation mechanism is speculative, accurate characterization of the interfacial oxide is an important s tep in avoiding its formation. Furthermore, characterization of the interfacial oxide provides information that could potentially be used to improve device reliability and performance. 4.3.3 Cameca APT Analysis Here, APT samples of the HEMT structures wer e prepared using FIB milling methods described in Section 3.1.4. The APT analysis on the Cameca system was performed at the National Institute of Materials Science in Tsukuba, Japan. Due to out of country location, only the Ni gate Si(111) and SiC sample s analyzed on this system. However, due to the previously successful field evaporation of Ni gate on Si(111) HEMTs, the work in this section focused on the Ni gate on SiC HEMTs.

PAGE 78

78 Many factors affected the quality of the field evaporation of the AlGaN/GaN HEMTs. Increasing specimen temperature increased the evaporation rate, but it did not reduce the frequency of the sample tip fracture. Decreasing laser power increased the time until specimen fracture, but it also did not prevent fracture. However, if the laser power was too large, the laser would damage the sample preventing evaporation. Concurrent scanning electron microscopy (SEM) during FIB atom probe tip fabrication reveals in Figure 4 6 A a specimen tip after complete fabrication and Figure 4 6 B a sample after fracture when using a laser power of 10mW. By comparing the thickness of the specimen (since all atom probe tips in this study are approximately the same geometry), the distance from the tip to the sample fracture can be estimated. Figur e 4 6 A shows that the tip has a diameter of approximately 1m about 10m from the apex of the tip. However, in Figure 4 6 B, the diameter of the tip at the fracture surface is approximately 2 to 4m indicating that the specimen fracture occurred more tha n 10m away from tip and area of interest. Thus, lowering the laser power prevents the catastrophic fracturing of tips from AlGaN/GaN HEMTs. Additionally, it is noted that adding a slight amount of He gas to the vacuum during evaporation also reduced the time until fracture for the specimen. The addition of He gas would reduce the magnitude of the electric field at the tip of the specimen. Thus, if there is a weak interface in the specimen, it may not fracture as quickly. Furthermore, the orientation of the specimen with respect to the tip apex was studied. Because fracture predominately occurred at the gate/AlGaN interface, different sample orientations were used to try to prevent fracture. By changing the orientation between the interfacial layer and the tip apex, the different layers (gate metal, interfacial

PAGE 79

79 layer, and AlGaN/GaN epilayers) are exposed at different and varying times during the evaporation. This affects the evaporation of the tip due to the change in electric field distribution at the tip apex due to the change in tip apex composition. For the sample orientation comparison, a transmission electron microscopy (TEM) micrograph of a representative cross section of a 1m gate length AFRL AlGaN/GaN HEMT is presented in Fig ure 4 7 A Here, high angle annular dark field scanning TEM (HAADF STEM) was used. The contrast mechanism correlates to the atomic number where brighter features correspond to areas of greater atomic number. In Fig ure 4 7 A the two bright layers of the gate are the Au a nd Ni layers while the two darker layers below them are the AlGaN and GaN epilayers. Additionally, Fig ure 4 7 A shows the directions of analysis for the four different orientations studied. The labels next to the direction of analysis in Fi gure 4 7 A cor respond to Fig ure 4 7 B E, respectively. The standard top down orientation where the sample tip is fabricated so that it matches the TEM cross section is shown in a pre sharpened state in Fig ure 4 7 B Here, the gate region is mounted above the AlGaN/Ga N epilayers and evaporation begins in the metal and progresses into the epilayers. The reverse top down orientation is shown in Fig ure 4 7 C in a pre sharpened state and is the opposite of the standard orientation. Here, the epilayers are mounted above t he gate so evaporation begins in the GaN and progresses into the gate metal. Additionally, it is noted that in some samples additional metal was deposited over the gate in order to increase the distance between the gate and the weld between the microsampl ed HEMT and W wire. The next orientation consists of a 90 rotation of the gate/AlGaN interface with respect to the cross section of the HEMT. In this orientation as shown in Fig ure 4 7 D evaporation

PAGE 80

80 occurs at the gate, interfacial layer, and AlGaN/GaN epilayers at the same time. The last orientation again consists of a 90 rotation of the gate/AlGaN interface but this time out of the plane of the cross section of the HEMT as shown in Fig ure 4 7 E Once again evaporation occurs at the gate, interfacia l layer, and AlGaN/GaN epilayers at the same time, but in this case evaporation could continue down the gate width of the device while in the previous orientation it would only be possible for the 1m across the gate length. From these four orientations, the standard top down orientation provided the best field evaporation before tip fracture. The reverse top down orientation exhibited poor evaporation and fractured. Additionally, the two 90 rotation orientations had the worst results with fractures occ urring immediately upon the initiation of field evaporation. These results indicate that including the interfacial layer in the immediate field evaporation resulted in early fracture of the tip, and may indicate that the interfacial layer is a weak featur e in the sample that contribute d to the samples constantly fracturing. probe run resulted from the standard top down orientation at 40K, 0.5 mW, and a small amount of He gas ad ded. This run showed part of the AlGaN epilayer and continued down into the GaN epilayer which is shown in Fig ure 4 8 Here, a slight amount of O is seen at the top of the AlGaN epilayer. This could be the bottom of part of the interfacial layer althoug h no other ions were detected in this region, or it is possibly some kind of contamination in the sample or from a fracture in the gate metal above the AlGaN. The majority of the other fractured tips only resulted in the evaporation of the Au layer of the gate followed by the beginning of the Ni evaporation, and then concluded with a sample

PAGE 81

81 tip fracture and evaporation of only GaN. Fig ure 4 8 shows the only run to contain part of the AlGaN epilayer. Thus, the interfacial layer between the gate and AlGaN epilayer and the top part of the AlGaN layer were not analyzed. 4.3.4 Aberration Corrected XTEM Chemical Analysis As discussed in the previous section, the native interfacial layer between the gate and AlGaN epilayer of the Ni gate on SiC HEMTs could not be analyzed due to incomplete field evaporation. There fore, high resolution aberration corrected chemical analysis of XTEM s from these samples was acquired. Figure 4 9 A shows a high magnification HAADF STEM image of the gate region and semiconductor interface of an unstressed Ni gate AlGaN/GaN HEMT on SiC. The distinct layers of the device are l abeled: Au, Ni, AlGaN, and GaN. Additionally, a highlighted box is shown in dicating the region that EELS maps were acquired from for chemical analysis. EELS maps of Ni, Ga, N, and O a re presented from the highlighted region of Figure 4 9 A in Figure 4 9 B, C, D, and E, respectively. The Ni EELS map in Figure 4 9 B shows the Ni layer of the gate and that it is flat and uniform. The Ga and N EELS maps s hown in Figure 4 9 C and D show the AlGaN layer below the gate. Comparing the Ni and Ga maps it is seen that there is a small gap between them. This is explained by analyzing the O EELS map in Figure 4 9 E which indicates that presence of an O layer between the Ni gate and AlGaN. This is comparable to the data shown in section 4.3.1 and 4.3.2 which indicates the interfacial layer between the gate and AlGaN was composed of an oxide. Thus, similarly to the Ni gate on Si(111), it is shown that the inte rfacial layer between the gate and AlGaN in the Ni gate on SiC is also composed of an oxide layer.

PAGE 82

82 4.4 Summary It was observed using XTEM that the there was an interfacial layer present between the gate and AlGaN of all AlGaN/GaN HEMTs analyzed in this wor k with varying thicknesses. However, it was only possible to analyze the interfacial layer of the Ni gate on Si(111) completely using APT. This successful APT field evaporation occurred on the Imago LEAP system which indicated that the interfacial layer was an oxide composed of two distinct regions. There was a NiO x layer at the bottom of the Ni layer of the gate, and there was an AlO x layer at the top of the AlGaN layer. Additionally, different APT systems and system parameters were tested in order to try to get successfu l evaporation from the other HEMT structures. However, the only factor that resulted in successful evaporation was when the interfacial layer between the gate and AlGaN epilayer was the thinnest. The samples with thicker interfacial layers did not successfully field evaporate the gate region, interfacial layer, and AlGaN epilayer together. Typically, either the samples fractured in the metal layer, or they fractured near the interfacial layer during evaporation and only the AlGaN or GaN layers were acquired. Thus, from both a reliability and characterization stand point the interfacial layer in AlGaN/GaN HEMTs should be made as thin as possible. Finally high resolution aberration correct TEM chemical analysis was used to identify th e interfacial layer of the Ni gate samples that could be field evaporated successfully. From the EELS maps it was concluded that there was an O signal between the Ni gate and AlGaN which indicated that in these devices the interfacial layer was an oxide, too.

PAGE 83

83 Figure 4 1 HAADF STEM images of the gate region of AlGaN/GaN HEMT s showing an increase in the interfacial layer thickness from left to right. A ) L ow magnification micrograph of a Ni gate on Si(111). B) Low magnification micrograph of a Ni gate on SiC. C) Low magnification micrograph of a Pt gate on SiC. D) H igh magnification mic rograph from the red box of part A E) H igh magnification micrograph from the red box of part B. F) H igh magnification micrograph from the red box of part C Parts reprinted with permission from M.R. Holzworth, N.G. Rudawski, S.J. Pearton, K.S. Jones, L. Lu, T.S. Kang, F. Ren, and J.W. Johnson, Characterization of the gate oxide of an AlGaN/GaN high electron mobility transistor, Applied Physics Letters. 98, 122103 (2011), Copyright 2011, American Institute of Physics

PAGE 84

84 Fi gure 4 2. Imago LEAP reconstructions of the gate/AlGaN interface showing the Ni/Au gate metal stack and AlGaN and GaN epilayers. A) Ni gate on Si(111) HEMT. B) Ni gate on SiC HEMT. C) Pt gate on SiC. Parts reprinted with permission from M.R. Holzworth, N.G. Rudawski, S.J. Pearton, K.S. Jones, L. Lu, T.S. Kang, F. Ren, and J.W. Johnson, Characterization of the gate oxide of an AlGaN/GaN high electron mobility transistor, Applied Physics Letters. 98, 122103 (2011), Copyright 2011, American Institute of Physics.

PAGE 85

85 Figure 4 3. Mass histories from field evap oration of the gate/AlGaN region of HEMTs using the Imago LEAP 3000X Si system A) Ni gate HEMT on Si(111). B) Ni gate HEMT on SiC. C) Pt gate HEMT on SiC

PAGE 86

86 Fig ure 4 4 P artial M/C spectra collected from APT analysis of the gate region o f an AlGaN/G aN HEMT structure on Si(111) Partial spectra are used to deduce a lack of Ga oxide in the interfacial oxide. A ) F rom M/C 40.5 to 44.0 indicating pea ks at 41.5, 42.0, and 42.5 B) F rom M/C 84.5 to 87.5 indicating no dis tinct peaks § § Reprinted with permission from M.R. Holzworth, N.G. Rudawski, S.J. Pearton, K.S. Jones, L. Lu, T.S. Kang, F. Ren, and J.W. Johnson, Characterization of the gate oxide of an AlGaN/GaN high elect ron mobility transistor, Applied Physics Letters. 98, 122103 (2011), Copyright 2011, American Institute of Physics.

PAGE 87

87 Fig ure 4 5 1D concentration profiles from the gate/AlGaN interface region of an AlGaN/GaN HEMT on Si(111). A ) The complete 1D atomic concentration profile for the gate region of the HEMT indicating an O peak at the Ni/AlGaN interface B ) 1D ionic concentration profil es of O + 27 AlO ++ and 58 NiO + measured in the vicinity of the interfacial layer indicating the oxide is composed of distinct AlO x and NiO x layers. ** ** Reprinted with permission from M.R. Holzworth, N.G. Rudawski, S.J. Pearton, K.S. Jones, L. Lu, T.S. Kang, F. Ren, and J.W. Johnson, Charact erization of the gate oxide of an AlGaN/GaN high electron mobility transistor, Applied Physics Letters. 98, 122103 (2011), Copyright 2011, American Institute of Physics.

PAGE 88

88 Figure 4 6 BSE micrographs of APT tips from AlGaN/GaN HEMTs. A) A freshly micromachined APT tip B ) T he same APT tip from part A after laser assisted field evaporation utilizing a laser power of 10mW.

PAGE 89

89 Fig ure 4 7 A XTEM image and BSE micrographs of the orientation of the gate and epilayers of an AlGaN/GaN HEMT. A) HAADF STEM micrograph showing the c ross section of a Ni gate AFRL AlGaN/GaN HEMT including the gate metal and AlGaN and GaN epilayers. The analysis direction for the different sampl e orientations is indicated. B ) BSE micrograph of a pre sharpened standard top d own oriented atom probe tip. C ) BSE micrograph of a pre sharpened reverse top down oriented atom probe tip. D ) BSE micrograph of a pre sharpened 90 rotated with respect to the HEMT cross section atom probe tip. E ) BSE micrograph of a pre sharpened 90 rotated out of the plane of the cross section atom probe tip.

PAGE 90

90 Fig ure 4 8 A p artial APT reconstruction showing part of the AlGaN epilayer and the GaN layer below it. Additionally, there is a layer of O above the AlGaN epilayer which may be part of the interfacial layer between the gate metal and AlGaN or it may be contamination from the metal fracturing above the AlGaN.

PAGE 91

91 Figure 4 9. XTEM image and EELS chemical maps of the gate/ AlGaN interface region from a Ni gate on SiC HEMT. A) HAADF STEM image of the gate/semiconductor interface with the distinct layers of the HEMT labeled and the box indicated for the EELS che mical analysis. B) Ni EELS map of the gate, interfacial la yer, and AlGaN. C) Ga EELS map of the gate, interfacial l ayer, and AlGaN. D) N EELS map of the gate, interfacial l ayer, and AlGaN. E) O EELS map of the gate, interfacial layer, and AlGaN indicating the interfacial layer is an oxide

PAGE 92

92 CHAPTER 5 EFFECT OF ELECTRIC FIELDS ON RELIABILILITY 5.1 Need to Understand Physical Defects in AlGaN/GaN HEMTs Prev ious work has reported the p hysical degradation of the AlGaN epilayer at the edges of the gate contact and interactions between the gate contact metal and epilayers have been observed using transmission electron microscopy (TEM), scanning electron microsco py (SEM), and atomic force microscopy. 21,62 64,99 This degradation occurred during electrical stressing. The epilayer degradation near the gate contact, degradation of the Ni/AlGaN interface, and transport of gate metal into the epilayers decreases drain current, which contributes to a reduction in the reliability of the HEMTs. 11 ,15,21 Understanding the mechanisms behind these interactions could enhance the reliability of A lGaN/GaN HEMTs by circumventing the defect formation conditions and preventing device degradation. Therefore, accurate structural characterization of defects is crucial in understanding the defect formation mechanism(s) and potentially improving HEMT reli ability and performance. Here, high angle annular dark field scanning TEM (HAADF STEM) was used to characterize a structural defect that forms between the Ni layer of a Ni/Au gate and the AlGaN epilayer during electrical stressing. It was found that the defect morphology approximates the simulated electric field during device electrical stressing indicating the shape of the field is an important factor influencing defect formation. Parts reprinted with permission from M.R. Holzworth, N.G. Rudawski, P.G. Whiting, S.J Pearton, K.S. Jones, L. Lu, T.S. Kang, F. Ren, E. Patrick, and M.E. Law, Field induced defect morphology in Ni gate AlGaN/GaN high electron mobility transistors, Applied Physics Letters. 103, 023503 (2013), Copyright 2013, American Institute of Physics.

PAGE 93

93 5.2 Experimental 5.2.1 Device Structure T he AlGaN/GaN HEMTs used for this work were all grown on the same semi insulating 6H SiC substrate and received the same processing. An AlN nucleation layer was used on the SiC and followed by a 2.25 m thick Fe doped GaN buffer, 15 nm of Al 0.28 Ga 0.72 N, and capped with 3 nm of an unintent ionally doped GaN layer (this layer is not normally visible in TEM micrographs). The ohmic contacts consisted of metal stacks of Ti/Al/Ni/Au annealed for 30 s at 850 C. 16,19 20 Additionally, the gate contact cons each gate having a 150 m width and a gate length ( L G ) of 100 nm (TEM micrographs indicate L G ~ 55 nm at the Ni/AlGaN interface). Lastly, the devices were passivated with SiN x A typical cross section of the devices is shown in the HAADF STEM image presented in Fig. 1, with the source, gate, and drain contacts indicated. 5.2.2 Electrical Stress Conditions Electrical data was measured using a HP 4156C semiconductor parameter analyzer. Off state high reverse gate bias conditions were used to electrically stress the HEMTs. During the applied stress, a constant drain source voltage V DS = 5 V was appli ed to the majority of the samples. Additionally, for some samples a V DS = 0, 10, or 15 V was applied However, the gate source voltage ( V GS ) was varied for each sample during stressing and started at either 5 or 10 V and ended at 42 V with the gate voltage stepped at 1 V/min. Electrical parameters of each HEMT were measured b etween each unit decrease in V GS at V DS = 5 V and V GS = 0 V. All stressing occurred at 25 C and in ambient atmosphere.

PAGE 94

94 5.3 Results and Discussion In order to better understand and correlate structural changes in HEMTs to performance changes shown by a ch ange in the output electrical curves, both electrical characterization and structural/chemical characterization will be discussed in the following sections. In order to make these connections clear, different electrical parameters and their change compare d to the physical change in the HEMT will be compared in separately. 5 .3.1 Relationship between I Dmax and Large Scale Physical Degradation The majority of HEMTs underwent an electrical stress with V DS = 5 V; therefore, this stress condition will be addres s first. This is will be followed by a discussion of representative curves from all V DS stress conditions. T he complete cross section of an unstressed HEMT is shown in Figure 5 1. This is shown to late r compare the image against degraded samples. After application of the electrical stress with V DS = 5V two normalized maximum drain current ( I D max ) of the HEMTs as given by I D max ( V GS )/ I D max (0), where I D max ( V GS ) is t he maximum drain current measured after the applied gate stress of V GS and I D max (0) is the maximum drain current of the unstressed device measured before the application of electrical stress. For each device, the applied electrical stress decreases the measured I D max ; therefore, the normalized I D max decreases as stressing continues. The first typical behavior of normalized I D max is I D slow degradation process. On av or show a minimal decrease of ~3 .5 % in normalized I D max by V GS = 22V. Furthermore, by the end of stressing at V GS = 42V, normalized I D max has decreased on average by

PAGE 95

95 ~13 4 %. This behavior is in contrast to the second behavior typically shown by the I D rapid degradation process. For these HEMTs, normalized I D max has decreased by ~16 % by V GS = 22 V, which is more than a 350 % larger decrease in normalized I D max compared V GS = 42 V, normalized I D max has decreased by ~26 %, which is a decrease in normalized I D max of 160 % compared to the cess. Lastly, it should be noted that the overall decrease in drain current ( I DS ) was permanent after stressing as revealed by the I DS versus V GS family of approximately 48 h after stressing. Furthermore, the permanent decrease in I DS was degradation sample at V GS = 2 V. Additionally, the gate current ( I G degradation HEMTs shown in Figure 5 I G slow dramatically increases by approximately two orders of magnitude at a lower V GS I G rapid HEMTs it is shown that only after the jump in I G at V GS = 14 V does the normalized I D max begin to degrade. Before the jump in I G I D max had degraded by <1 % over the first 10 V of applied gate stress; however, after the I G jump, I D max begins to degra de slowly and constantly at ~0.5 %/V until the end of stressing. In contrast, for the HEMTs that I D max begins to decrease as soon as stressing initiates. This contrast between the characteristic I D max and I G curves of th e sets of devices indicates that different degradation modes are occurring.

PAGE 96

96 Following the application of electrical stress, the HEMTs were serially sectioned along the width of the gate using site specific FIB milling. Each sample was serially sectioned a t approximately 9 15 locations with the sections approximately 10 m long by 1 m wide by 2 m deep and spaced approximately 10 15 m apart. From the exposed cross sections, the gates were inspected for defects and irregularities using cross sectional scanning electron microscopy (XSEM) at an incident angle of 52. After examining the gate/AlGaN epilayer interface using XSEM, some sections were further thinned for TEM analysis. HAADF degradation samples are p resented in Figure 5 3 A and Figure 5 3 B, respectively. In these images, brighter features correspond to areas of greater average atomic number; the individual Au, Ni, AlGaN, and GaN layers are indicated in Figure 5 3 A. own in Figure 5 3 A, shows no obvious physical degradation and the gate contact has maintained the same morphology as the unstressed control HEMT shown at lower magnification in Figure 5 1. Furthermore, no egradation samples. However, the shown in Figure 5 shaped defect under the gate, which penetrates into the AlGaN epi layer at two points. Although this defect penetrates the AlGaN epilayer, it stops before crossing the AlGaN/GaN interface. However, the presence of this defect near the interface may still impact the electrical performance of the device possibly by influ encing leakage and/or trapping, and it may explain the rapid decrease in normalized I D max that exhibited no early, rapid decrease in normalized I D max and no observed defect

PAGE 97

97 formation. It should be noted that this d efect was present throughout the entire device width and corresponds to the I D rapid and I G rapid curves shown in Figure 5 2 indicating the largest initial drop in normalized I D max Additionally, I G degradation HEMTs showing no p degradation HEMTs, which contain physical defects indicating that I D max is a more useful measure of physical device degradation. Furthermore, this pattern of an initial degradation in I D max has been ob served in multiple devices containing physical degradation HEMTs, indicating that th e defect formation pushed the gate vertically from its initial position and partially separated it from the SiN x passivation. Thus, it appears that a rapid decrease in I D max is possibly correlated to large physical defect formation under the gate. In add ition to HEMTs stressed using a V DS = 5 V, other HEMTs where stressed with V DS varying from 0 V to 15 V in increments of 5 V. The change in normalized I D max against V DS for representative devices from these stress conditions are shown in Figure 5 4. Aga in, it can be seen that from the ~30 devices stressed in total, that two I D max degradation cases shown with V DS = 5 and 10 V, the decreases immediately from the start of electri cases shown with V DS = 0, 5, and 15 V that the I D max The normalized I D max and I G V DS = 0, 5, and 15 V are p lotted in Figure 5 5, Figure 5 6, and Figure 5 7, respectively. It is clear from these

PAGE 98

98 figures that once again the I D max does not decrease until the I G jump occurs. Thus, it I D max is linked to the I G jump. It is interesting to note that the slopes of the I D ma x degradation I G jump ~0.5 %/ V I t is noted that even during sections of greatest I D max HEMTs that the average decrease is still ~0.5 %/V. This may indicate that these devices degrade by similar and constant mechanisms after the I G jump possibly due to the creation of traps from the off state electrical stressing. Thi s is in contrast to the the slopes of their overall I D max degradation are ~ 0.8 1 %/V, and during sections of greatest I D max degradation range from ~ 1 2 %/V. Following the electrical stressing at V DS = 0, 5, 10, and 15 V, the HEM Ts were again serially sectioned inspected by XSEM and fabricated into XTEM samples as discussed previously. HAADF samples at V DS = 5 V were previously presented in Figure 5 3 A and Figure 5 3 B, res pectively. V DS = 0 and 15 V and the rapid degradation same at V DS = 10 V are presented in Figure 5 8 A C, respectively. Again, i n these HAADF STEM images, brighter features correspond to areas of greate r average atomic number; the individual Au, Ni, AlGaN, and GaN layers are indicated in Figure 5 8 A. Again, t s, shown in Figure 5 8 A and B, show no obvious physical degradation and the gate contact has maintained the same morph ology as the unstressed control HEMT shown at lower magnification in Figure 5 1 3 A Furthermore, no defects were

PAGE 99

99 ever observed in any of the se at V DS = 10 V exhibits a markedly different m orphology as shown in Figure 5 8 C This image in dicates the presence of a shaped defect under the gate, which penetrates into the AlGaN epilayer Although this defect penetrates the AlGaN epilayer, it stops before crossing the AlGaN/GaN interface. 3 B. This aerial difference probably helps contribute to the larger dro p in I D max at 26 % as shown in the I D max plot in Figure 5 4. Additionally, energy dispersive X ray spectroscopy (EDS) and electron energy loss spectroscopy (EELS) analysis were performed on the shaped shaped defect s suggesting the defect region s contain Al, Ga, Ni, and O but not N. 9 A, C, and E respectively. Additionally, the position of the line scans is indicated by the lines on the HAADF STEM micrographs on Figure 5 9 A, C, and E. The EDS line scan for the unstressed HEMT is shown in Figure 5 9 B from the analysis region indicated in Fig ure 5 9 A. The EDS line scan indicates that there are four distinct compositional areas of the gate/semiconductor interface region: the Au, Ni, AlGaN, GaN layers. These original, distinct layers are indicated on all three of the EDS line scans. It is noted that there is no diffusion or reaction indicated between the gate metals and the AlGaN layer in this Figure 5 9 D from the region indicated by Figure 5 9 C Her e, the EDS line scan

PAGE 100

100 clearly indicates that there defect. This implies that there is some mixing of the Ni layer of the gate and the AlGaN epilayer. However, the signal quality from the line sca n drops in the defect in the AlGaN layer. This may be due to two reasons. First, the defect may be of lower density due to the reaction between the layers and thus less material may be present to produce signal for analysis. Second, the combination of t he NI, Al, and Ga may create a matrix effect that emits very few X rays. is shown in Figure 5 9 F from the region indicated in Figure 5 9 E. Similar to the that Ni, Al, and Ga are again present throughout the defect. This indicates that there appears to be a mixing of the Ni, Al, and Ga defects appear to have roughly th e same cation chemistry for their composition. Additionally, the mixing of the Ni, Al, and Ga is shown by the decrease in the Ni signal unstressed HEMT, the Ni signal is the large st of the signals; however, in the defects it is clear that the Ni signal is drastically reduced. This indicates a decrease in Ni concentration in the original Ni region of the gate presumably due to its mixing into the AlGaN epilayer. In addition to ED on the same image from each defect in order to better show the change in composition. Additionally, the maps are shown in partial transparency and opa que in help view the transition of the composition. Figure 5 10 A shows a bright field TEM (BF

PAGE 101

101 background image used for comparison to the EDS maps of this defect. Figure 5 10 B and C show the Ni maps of th and no transparency respectively. These images indicate that there is a large Ni signal present around the original Ni layer of the gate and that there is a small Ni signal present in the defect. This small signal from the d efect could again be due to the change in composition. Figure 5 respectively These maps clearly show that Ga has diffused into the defect near the gate region and that the Ga signal is lower in the AlGaN epilayer where the defect is present. Additionally, Al maps with partial and no transparency are respectively shown in Figure 5 10 F and G. Like the Ga maps, the Al maps show that Al has diffused into the defect near the gate region These results match well with the previous EDS line defect indicating that the defect is a mixture of Ni, Ga, and Al. and opaque in order t o help view the transition of the composition on the same image. Figure 5 11 A shows a HAADF background template used for comparison to the EDS maps for this defect. The EDS maps of Ni for the wave defect are shown in Figure 5 11 B and C in partial and no transparency respectively In these maps, Ni is shown to be present just below the Au region of the gate where the original Ni layer of the gate was and it extends down into the top half of the defect. There is some disagreement here between the Ni EDS line scans and maps as in the line scans Ni is shown to be throughout the defect, but in the maps it is depicted as being in the top half of the defect. However, this could simply be due to the sample s ize, data acquisition time, sample drift, and resolution of the map.

PAGE 102

102 Ga maps of the shown in Figure 5 11 D and E with partial and no transparency, respectively. From these figures, it is clear that Ga has diffused up into the defect. Additionally, the Al maps are presented in Figure 5 11 F and G with partial and no transparency, respectively. No conclusion is drawn from these figures as there is too much Al signal noise in the maps which is indicated by the AlGaN not being sharp in resolution. However, the Ni and Ga maps can be compared to the EDS line scans which agree that the defect is composed of a mixture of Ni and Ga and possibly Al. While EDS is an ideal analysis tool for larger Z elements and metals it has dif ficulty det ec ting low Z elements. Thus, the anion groups for the new composition of erefore EELS mapping was used in order to view the complete composition of these defects. Similar to the EDS maps the EELS maps are overlaid on the same image from each defect in partial and no transparency in order to better s how the change in composition. defect the background template BF TEM image is shown in Figure 5 12 A. Ni EELS maps were acquired transparency in Figure 5 12 B and C, respectively. From these figures, Ni is shown to be present just below the Au region of the gate at the original Ni layer of the gate, at the bottom of the defect in the AlGaN epilayer showing the interface between the defect and epilayer, and in some regions between the gate and defect/AlGaN interface. This verifies the EDS line scan data showing the Ni present throughout the defect that was shown f 5 9 D Next, the O EELS maps are presented in partial and no transparency in Figure 5 12 D and E, respectively. Inspection of these figures shows there is an increased signal at the side of the defect at the defect/SiN x

PAGE 103

103 interface and at the bottom of the defect at the defect/AlGaN interface. These figures may indicate that O is present in the defect. Lastly, the N EELS maps in partial and no transparency are shown in Figure 5 12 F and G. Clearly from these figures, there is shown to be a deficiency of N in the defect at both the original Ni layer of the gate and where the defect penetrates the Al GaN. Therefore, these EELS maps indicate there is some measured Ni throughout the defect, and that the defect is shown to contain O and lack N where it penetrates the AlGaN epilayer indicating that probably an oxidation reaction has occurred. Now, f S maps are presented as partially transparent and opaque in order to help view the transition of the com position on the same image. Figure 5 13 A shows a HAADF background templat e used for comparison to the EELS maps for this defect. Ni EELS maps are presented in partial and no transparency in Figure 5 13 B and C, respectively. Similar to the EDS maps, the EELS maps show Ni present just below the Au region of the gate where the original Ni layer of the gate was and extending down into the top half of the defect. Next, the O EELS maps are shown in partial and no transparency in Figure 5 13 D and E, respectively. It is clear from these figures that O is present in the original Ni layer of the gate, and there is an increase in O signal in the defect itself particularly in the original AlGaN region as the O signal mimics the shape of the defect. Lastly, the N EELS maps are presented in partial and no transparency in Figure 5 13 F and G, respectively. Here, N is shown to be deficient in the defect at both the original Ni layer of the gate and where the de fect penetrates into the AlGaN. Therefore, these elemental maps indicate there is some measured diffusion of Ni into

PAGE 104

104 the defect and that the defect is shown to contain O and lack N where it penetrates the AlGaN epilayer indicating that an oxidation reacti on has occurred. Additionally, it exhibit this NI, Ga, and Al mixing in the defect with O and the lack of N. T herefore, it is Al Ga oxide. This EDS and EELS work is consistent with the HAADF STEM image of the Figure 5 3 B and Figure 5 8 C in which the contrast mechanism arises from differences in the a verage atomic number. In general, the defect s are darker in contrast than both the Ni and the AlGaN epilayer. This implies that there was a decrease in average atomic number, which is consistent with the EDS and EELS results showing the presence of O in the defect. In order to better understand the change in shape of the gate upon degradation, the Florida Object Oriented Device Simulator. 100 101 Measured electrical data was used to calibrate the simulated results to actual device performance, and no temperature effects were taken into account in the simulation due to the off state stress where there should be negligible curren t flow and heating. For the V DS = 5 V stress, a t approximately V GS = 23 V, I G rapid has jumped two orders of magnitude, suggesting an electronic or physical leakage path between the two dimensional electron gas at the AlGaN/GaN interface and gate as show n in Figure 5 2; however, I D rapid has decreased shaped defect shown in Figure 5 3 B could have formed continuously throughout the entire application of

PAGE 105

105 electrical stress. This same analogy i s possible for the V DS = 10 V Consequently, the electric field for the V DS = 5 V stress condition is plotted at the final voltage condition, V DS = 5 V and V GS = 42 V, and is shown in Figure 5 14 where iso c ontour lines of the magnitude of the vertical and lateral components of the electric field are indicated. It is noted that due to the inverse piezoelectric effect, the electric and strain fields are linked. 102 The refore, the magnitude of the strain field is proportional to this simulated electric field at these bias conditions and would possess the same shape. From the simulations, it appears the 6 MV/cm sha ped defect Additiionally, t he Ni gate structures exhibiting gate edge pitting by Gao et al. which are much larger than the work presented here, were simulated. It is mentioned again that pitting ~60 85 nm in diameter was observed. Here, the simulati ons of the electric field contours for the Gao et al. HEMTs are plotted in Figure 5 15 A B at the stress state of V DS = 0 V and V G S = 4 0 V. Furthermore, from figure 5 15 B it is observed that the 5 6 MV/cm contour line s approximately models the s ize of the observed gate edge pitting by Gao et al. This simulation result paired with the 14 may indicate that the electric/strain field may influence the evolution of defect morphology. Previous work has suggested that defects and reactions occur where the electric field is largest in a HEMT, at the gate edges. 61,64 However, in this case, due to this large, nearly symmetric electric/strain field present, it is poss ible that the fields influence the diffusion or reactivity of the layers and O directly contributing to the morphology of the defect and explain the similarity between the defect shape and the simulated electric/strain field contours. This could

PAGE 106

1 06 indicate that it is the shape of the electric/strain field itself that influences the defect morphology and not just the large magnitude of the field for HEMTs with L G < 100 nm As previously mentioned the O in the defects most likely came from the outside ambient during the stressing due to the similar processing each HEMT received, and each HEMT was on the same wafer in near proximity to each other Due to the processing technique used to deposit SiN x in these devices there are air gaps in the SiN x passivatio n. F rom the edge of these gaps, channel s runs to the gate edges. This would provide an easier diffusion path for O to reach the gate edges where it could readily react with the AlGaN or Ni. Figure 5 16 A show s a low magnification BF TEM image of the g ate region of a HEMT showing the air gaps in the SiN x passivation on both sides of the gate. The higher magnification BF TEM image shown in Figure 5 16 B indicates that channels running from the gaps to the gate edges. Thus, these defects could be the result of an oxidation due to the low barrier diffusion pathway between the gate edge and the ambient. 5.3.2 Relationship between I G and Large Scale Physical Degradation As shown in Figure 5 2 and Figure 5 4, w hen a 100 nm gate HEMT is electrically st ressed, it usually exhibits a sharp jump in the I G which is also known as the critical voltage. Thus, the I G jump was investigated to determine whether or not it was linked to any physical degradation in the devices. HEMTs were stressed to three specif ic conditions: before the I G jump, at the I G jump, and after the I G jump. These electrical conditions comparing the I G to V GS the can be seen in Figure 5 17 wher e the representative of the before I G jump, epresenta tive of during the I G jump, and representative after the I G jump condition.

PAGE 107

107 Following the application of electrical stress, the HEMTs were serially sectioned along the width of the gate using site specific FIB milling. Each sample w as serially sectioned at approximately 9 15 locations with the sections approximately 10 m long by 1 m wide by 2 m deep and spaced approximately 10 15 m apart. From the exposed cross sections, the gates were inspected for defects and irr egularitie s using XSEM at an incident angle of 52. After examining the gate/AlGaN epilayer interface using XSEM, some sections were further thinned for TEM analysis. Representative XTEM samples analyzed by HAADF STEM of a HEMT stressed before the I G jump, a HEMT stressed until it was within the I G jump, and a HEMT stressed past the I G jump are sh own in Figure 5 18 A C respectively. It is cle a r from the images in Figure 5 18 that no large physical degradation has occurred at the gate. Additionally, it is noted that no large physical degradation was ever noted for these samples Thus, it appears that the dramatic change in I G is not an indicator of large physical degradation in AlGaN/GaN HEMTs. 5.4 Summary High angle annular dark field scanning transmission elec tron microscopy was used to study a reaction based defect present between the Ni layer of a Ni/Au gate and the AlGaN epilayer of an AlGaN/GaN high electron mobility transistor. For devices stressed at V DS = 5 V where the normalized maximum drain current w as rapidly and shaped oxidation defects associated with significant reaction between the Ni and AlGaN were observed. This is particularly noticeable during early electrical stressing where the d ifference between the two sets of devices was greatest. Additionally, when the sample set is expanded to include V DS stress ranging from 0 to 15 V, this trend of a rapid decrease in I D max

PAGE 108

108 associated with physical degradation continues. For example, at V DS shape oxidative defect was found between the gate and into the AlGaN epilayer. Due to the devices originating from the same wafer with identical processing, the O in the defect likely could have come from the SiN x passivation or from t he ambient after diffusion through the passivation. This low barrier to diffusion could be due to the channels present in the SiN x passivation between the gate edges and the air gaps in the SiN x which lead to the outside ambient. Additionally, the defect imitates the shape of the simulated electric/strain field iso contour at approximately 6 MV/cm present in the device during electrical stressing ilar behavior between 5 6 MV/cm. Furthermore, simulations of the gate structure and bias conditions from Gao et al. reveals the 5 6 MV/cm contours matches the size of the pitting, too. This suggests that the shape of the field may influence defect morphology However, this electric field condition m ay not be suf ficient to generate defects as there may be multiple failure differences in degradation Finally the jump in I G was compared to gate defect formation. It was shown that whether or not the samples were stressed to before the I G jump, until with the I G jump, or past the I G jump that no physical deformation was detected. This indicates that relying on I G measurements as a indicator for physical degradation is a poor choice while the I D max during stressing.

PAGE 109

109 Figure 5 1. L ow magnification HAADF STEM image of a cross section of an unstressed Ni gate AFRL AlGaN/GaN HEMT on SiC with the source (S), gate (G), and drain (D) contacts labeled Reprinted with permission from M.R. Holzworth, N.G. Rudawski, P.G. Whiting, S.J. Pearton, K.S. Jones, L. Lu, T.S. Kang, F. Ren, E. Patrick, and M.E. Law, Field induced defect morphology in Ni gate AlGaN/GaN high electron mobility transistors, Applied Phy sics Letters. 103, 023503 (2013), Copyright 2013, American Institute of Physics.

PAGE 110

110 Figure 5 2. M easured normalized I D max and I G degradation HEMTs when stressed from V GS = 5 or 10 V to 42V at 1 V/min and V DS maintained at 5 V throughout stressing §§ §§ Reprinted with permission from M.R. Holzworth, N.G. Rudawski, P.G. Whiting, S.J. Pearton, K.S. Jones, L. Lu, T.S. Kang, F. Ren, E. Patrick, and M.E. Law, Field induced defect morphology in Ni gate AlGaN/GaN high electron mobility transistors, Applied Physics Letters. 103, 023503 (2013), Copyright 2013, American Institute of Physics.

PAGE 111

111 Figure 5 3 H igh magnification HAADF STEM images of the gate region of AlGaN/GaN HEMTs showing the distinct Ni/Au gate metal stack layers a nd AlGaN and GaN epilayers A ) HEMT experienci basically indistinguishab le from an unstressed de vice. B ) HEMT shaped defect formed from the original Ni layer of the gate and penetrating into the AlGaN epilayer. *** *** Reprinted with permission from M.R. Holzworth, N.G. Rudawski, P.G. Whiting, S.J. Pearton, K.S. Jones, L. Lu, T.S. Kang, F. Ren, E. Patrick, and M.E. Law, Field induced defect morphology in Ni gate AlGaN/GaN high electron mobility transistors, Applied Physics Letters. 103, 023503 (2013), Copyright 2013, American Institute of Physics.

PAGE 112

112 Figure 5 4 M easured normalized I D max representative curves of degradation HEMTs when stressed from V GS = 5 or 10 V to 42V at 1 V/min while V DS maintained at either 0, 5, 10, or 1 5 V throughout stressing

PAGE 113

113 Figure 5 5 Measured normalized I D max and I G representative curves of degradation AlGaN/GaN HEM T stressed from V GS = 10 V to 42V at 1 V/min while V DS was maintained at 0 V throughout stressing.

PAGE 114

114 Figure 5 6 Measured normalized I D max and I G representative curves of degradation AlGaN/GaN HEMT stressed from V GS = 5 V to 42V at 1 V/m in while V DS was maintained at 10 V throughout stressing.

PAGE 115

115 Figure 5 7 Measured normalized I D max and I G representative curves of degradation AlGaN/GaN HEMT stressed from V GS = 10 V to 42V at 1 V/min while V DS is maintained at 15 V through out stressing.

PAGE 116

116 Figure 5 8 High magnification HAADF STEM images of the gate region of AlGaN/GaN HEMTs showing the distinct Ni/Au gate metal stack layers and AlGaN and GaN epilayers. A) HEMT with V DS = 0 V which are ba sically indistinguishab le from unstressed devices. B) HEMT V DS = 15 V C ) HEMT experiencing with V DS = 10 V wave shaped defect formed from the original Ni layer of the gate and penetra ting into the AlGaN epilayer.

PAGE 117

117 Figure 5 9 High magnification HAADF STEM images of the gate region of AlGaN/GaN HEMTs and their corresponding EDS line scans. A) Representative HAADF STEM image of an unstressed HEMT. B) EDS line scan showing the four d istinct regions of the the unstressed HEMT: GaN, AlGaN, Ni, and Au. C) Representative HAADF STEM image of a HEMT shaped defect formed from the original Ni layer of the gate and penetrating into the AlGaN epilayer at V DS = 5 V. D) EDS line scan showing the lack of change in the Au and GaN layers but the mixing of the Ni and AlGaN. E) Representative HAADF STEM images of a HEMT shaped defect formed from the original Ni layer of the gate and penetrating into the AlGaN epilayer at V DS = 10 V. F) EDS line scan showing the lack of change in the Au and GaN layers but a mixing of the Ni and AlGaN layers.

PAGE 118

118 Figure 5 10 High magnification BF TEM images of the ga overlaid with EDS maps. A) Template image used for EDS map overlay for comparison. B) Ni EDS map shown in partial transparency. C) Ni EDS map shown with no transparency. D ) Ga EDS map shown in partial transparency. E) Ga EDS map shown with no transparency. F ) Al EDS map shown in partial transparency. G) Al EDS map shown with no transparency.

PAGE 119

119 Figure 5 11 High magnification HAADF defect overlaid with EDS maps. A) Templa te image used for EDS map overlay for comparison. B) Ni EDS map shown in partial transparency. C) Ni EDS map shown with no transparency. D ) Ga EDS map shown in partial transparency. E) Ga EDS map shown with no transparency. F ) Al EDS map shown in part ial transparency. G) Al EDS map shown with no transparency.

PAGE 120

120 Figure 5 12 High magnification BF overlaid with EELS maps. A) Template image used for EELS map overlay for comparison. B) Ni EELS map s hown in partial transparency. C) Ni EELS map shown with no transparency. D ) O EELS map shown in partial transparency. E) O EELS map shown with no transparency. F ) N EELS map shown in partial transparency. G) N EELS map shown with no transparency.

PAGE 121

121 Figure 5 13 High magnification HAADF defect overlaid with EELS maps. A) Template image used for EELS map overlay for comparison. B) Ni EELS map shown in partial transparency. C) Ni EELS map shown with no t ransparency. D ) O EELS map shown in partial transparency. E) O EELS map shown with no transparency. F ) N EELS map shown in partial transparency. G) N EELS map shown with no transparency.

PAGE 122

122 Figure 5 1 4. S imulation showing the iso contour lines of the magnitude of the vertical stress state of V DS = 5 V and V GS = 42V Reprinted with permission from M.R. Holzworth, N.G. Rudawski, P.G. Whiting, S.J. Pearton, K.S. Jones, L. Lu, T.S. Kang, F. Ren, E. Patrick, and M.E. Law, Field induced defect morphology in Ni gate AlGaN/GaN high electron mobility transistors, Applied Physics Letters. 10 3, 023503 (2013), Copyright 2013, American Institute of Physics.

PAGE 123

123 Figure 5 15. S imulation showing the iso contour lines of the magnitude of the vertical and lateral components of the electric field from Ni gate HEMTs structures by Gao et al. A) Low magnification simulation image of the entire gate at a stress state of V DS = 0 V and V GS = 40 V B) High magnification of the simulation image from A) indicated by the b lack box where the pitting defects Gao et al. observed are approximately the same size as the 5 6 MV/cm contours.

PAGE 124

124 Figure 5 16. Representative BF TEM images show how th e air gaps in the SiN x passivation lead to ambient and the channels connecting th e gate edges to the gaps. A) Low magnification image showing the air gaps with respect to the gate with the orange box indicating the region for higher magnification analysis in part B. B) High magnification image showing the channel that runs between th e gate edges and gaps possibly allowing easier diffusion of O

PAGE 125

125 Figure 5 17 Measured I G representative curves of stressed AFRL AlGaN/GaN HEMTs on SiC demonstrating the before, middle, and after I G jump characteristics. HEMTs were stressed from V GS = 10 V up to 33 V at 1 V/min while V DS was maintained at 5 V throughout stressing

PAGE 126

126 Figure 5 18 High magnification HAADF STEM images of the gate region of AFRL AlGaN/GaN HEMTs on SiC showing the distinct Ni/Au gate metal stack layers and AlGaN and GaN epilayers A) Representative HEMT stressed before the I G jump. B) Representative HEMT stressed to with in the I G jump C) Representative HEMT stressed past the I G jump.

PAGE 127

127 CHAPTER 6 EFFECT OF THERMAL STRESS ON RELIABILITY 6 .1 I mportance of Understanding the Thermal Exposure to HEMTs Normally thermal annealing is used to accelerate failure of AlGaN/GaN HEMTs in conjunction with electrical biasing. Due to this dual application of str ess, failure mechanisms caused only by the rmal or electrical stress may not be detected. However, normal on state operating conditions with small electrical stress can produce internal temperatures up to 300C. Additionally, high temperature storage and applications of HEMTs require an understanding of how pure thermal stress can modify the device structure and electrical properties of the HEMTs in order to improve reliability and device life time models. Here, AlGaN/GaN HEMTs on SiC and Si(111) substrat es with no electrical or mechanical stress are subjected to only thermal stress using anneals ranging from 300C to 850C for duration s of 0.5 h to 400 h in order to isolate the effects of temperature on HEMT degradation. 6 .2 Experimental The re are two typ es of thermal anneals presented in this work First, there are the iso chronal anneals which are anneals at different temperatures but for the same duration. Second, there are iso thermal anneals which are anneals at the same temperature but for varying duration For clarity, the descriptions of the specific anneals are broken down in the following sections. 6 .2.1 Iso chronal Anneals Iso chronal anneals were completed on HEMTs on both SiC and Si(111) substrates. For the AlGaN/GaN HEMTs grown on SiC use d for this work, all were grown on the same semi insulating 6H SiC substrate and received the same

PAGE 128

128 processing. An AlN nucleation layer was used on the SiC and followed by a 2.25 m thick Fe doped GaN buffer, 15 nm of Al 0.28 Ga 0.72 N, and capped with 3 nm of an unintentionally doped GaN layer (this layer is not normally visible in TEM micrographs). The ohmic contacts consisted of metal stacks of Ti/Al/Ni/Au annealed for 30 s at 850 C. 16,19 20 Additionally, the gate contact consisted of a Ni/Au metal stack. The devices 150 m width and a gate L G = ~100 nm or ~1 m. For the AlGaN/GaN HEMTs grown on Si(111), all were grown on the same Si substrate and received t he same processing which was nearly identical to the HEMTs grown on SiC. Again, HEMTs on Si(111) had a 2.25 m thick Fe doped GaN buffer, 15 nm of Al 0.28 Ga 0.72 N, and capped with 3 nm of an unintentionally doped GaN layer (which is not normally visible). The ohmic contacts consisted of metal stacks of Ti/Al/Ni/Au annealed for 30 s at 850 C. 16,19 20 Additionally, the gate contact consisted of a Ni/Au ch gate having a 150 m width and a gate length L G = ~100 nm or ~1 m Devices on SiC were annealed for 0.5 h at 450C, 650C, 750C, and 850C. Each HEMT was only exposed to one temperature in order to isolate the effects of that temperature on structural changes. Devices on Si(111) were annealed for 0.5 h at 300C, 350C, 400C, 450C 500C, 550C, and 600C 6 .2.2 Iso thermal Anneals Iso thermal anneals were performed only on HEMTs on Si(111) substrates. The same device structure for the Si(111) HEMT s was used in the iso chronal anneals. The HEMTs were placed in a furnace at 300 C and annealed for varying durations. Samples were removed at 0.5 h 1 h 2 h 4 h 8 h 24 h 48 h 96h, 192 h, and 400 h.

PAGE 129

129 6 .3 Results and Discussion 6 .3.1 Iso chronal Anneals on SiC HEMTs Iso chronal anneals on SiC HEMTs were undertaken to show the difference between defects created electrically as presented in Ch. 5 and thermally in sub m gate length HEMTs. After subjecting the HEMTs on SiC to thermal anneals, XTEM s amples were prepared using the FIB. BF TEM of a pre and two post annealed sub m gate HEMT is shown in Figure 6 1 A C respectively. Comparing the two images it is clear that the Ni layer of the gate has been modified. Additionally, it is clear from t he pre and two post annealed HAADF STEM images in Figure 6 2 A C respectively that only the Ni layer of the gate is affected by the thermal anneal. However, it is not clear about the amount of interdiffusion present in the gate contact as the modifica tion of the Ni layer prevents this analysis. T he post annealed HEMTs were also subjected to the etching described earlier in section 3.1.2. PSEM of the sub m and 1 m gates was used to characterize the top surface of the AlGaN epilayer of the HEMTs. F ig ure 6 3 A shows the PSEM of an unannealed sub m HEMT which shows that the top surface is pristine of any defects. For the thermally annealed samples, two distinct types of surface features were revealed by PSEM First, no surface change is visible as shown in Fig ure 6 3 B Second, a surface reaction has occurred as revealed by the etching of the AlGaN layer as shown in Fig ure 6 3 C Additionally, from the PSEM shown in Figure 6 3 it was shown that when a surface reaction does occur that it is rare and only consumes a small percentage of the gate. The exact mechanism of this reaction is unknown as it was never captured in a XTEM sample due to its low density of formation unlike the

PAGE 130

130 electrical defects which when formed always possessed more than 30% of the gate width. For the 1 m gate HEMTs, PSEM performed on etched samples revealed two types of surfaces, too. For comparison to the annealed devices, an unannealed HEMT surface is shown in Fig ure 6 4 A Figure 6 4 B shows an annealed sample at 450C wh ich demonstrates that no change in the surface was observed. However, Fig ure 6 4 C shows some etch pits on the surface of a sample exposed to a 650C anneal. Additionally, Fig ure 6 4 D shows more etch pits on the surface of the etched HEMT as the tempera ture is increased to 750 C. As can be observed from Figure 6 4 C and D, these etch pits only formed under the gate contact and not under the SiN x passivation. it is assumed that O diffusion did not play a role in formation. Therefore, these pits are likely formed by the diffusion of gate metal down the threading dislocations present in epilayers of the device structure. Another indicator that this is likely a diffus ion based event is the increase in density of the pitting with increasing temperature. Furthermore, the size of the etch pits increased with temperature demonstrating that a larger thermal energy increased the pitting reaction rate. It is noted that high magnification PSEM of the etch pits shown in Figure 6 5 reveals their hexagonal symmetry which matches the symmetry of the epilayers. Due to this matching symmetry, the pits are unlikely to be due to a non equilibrium formation process indicating that this reaction could be easily modeled. 6 .3.2 Iso chrona l Anneals on Si(111) HEMTs Only 1 m gate length HEMTs were analyzed for the iso chronal anneals of HEMTs on Si(111). The purpose of these anneals was to determine the effect of

PAGE 131

131 temperature on the str uctural change of the gate metal. Similar to the iso chronal anneals on SiC HEMTs, XTEM samples of post annealed Si(111) HEMTs were prepared using the FIB. It was found that annealing between 300C and 550C resulted in gate metal entering into threading dislocations in the AlGaN epilayer. This is demonstrated in the low magnification HAADF STEM image Figure 6 6 A where gate metal can be observed penetrating into the dislocations. This penetration is more easily observed in the high magnification HAADF STEM image Figure 6 6 B. Additionally, it was observed that for moderate temperature anneals (>550C) even for a short anneal time of 0.5 h that delamination of the gate metal occurred as observed in Figure 6 6 C. While delamination of the gate metal i s a serious reliability issue that needs to be avoided, it prevented the metal from penetrating into the threading dislocation in the HEMT This was due to the metal now elevated above the surface of the AlGaN and not in connect ion with it or the dislocat ions anymore Thus, metal penetration into dislocation s was not observed using XTEM for temperature s > 55 0C. Additionally, anneals were capped at 0.5 h for comparison between the rates of metal penetration between t he anneal temperatures. This condition was applied because i t was observed that metal penetration saturated and stopped after 0.5 hr in anneal temperatures greater than 500C. This is demonstrated comparing high magnification HAADF STEM images of anneals at 500C for 0.5 h and 6 h in Fig ure 6 7 A and B respectively. It is clear from these images that the metal penetration has effectively stopped This conclusion is reached due to the penetration depth being approximately the same between the 0.5 and 6 h samples even though the anneal

PAGE 132

132 duratio n was 12 times longer between the samples Thus, the interdiffusion of the Ni Au layers in the gate contact limits the metal penetration At higher temperatures, the interdiffusion of the Ni Au is comparatively much larger than at lower temperatures. T hus t he concentration of Au into the metal penetration at higher temperatures is much greater than at lower temperature s A high magnification HAADF STEM image of gate penetration from a 0.5 h 500C sample is shown in Figure 6 8 A. An EDS line scan is shown through this defect in Figure 6 8 A indicated by the red arrow and is presented in Figure 6 8 B. The line scan shows that the Au composition in the defect is much larger than the Ni composition. This indicates that under this thermal stress co ndition, the defect has become mostly Au due to high interdiffusion between the Ni and Au. A quick study was completed to view the effect of gate metal on penetration depth. AlGaN/GaN HEMTs were etching following the procedures from section 3.1.2. These HEMTs were then separated into two groups where one group had only Ni deposited and the o ther had only Au deposited on the now exposed AlGaN surface Following metal deposition, the metal was passivated with PECVD SiN x This effectively made two sets of samples with HEMTs having only Ni or Au gates instead of the original Ni/Au structure. The single gate metal HEMTs were then annealed for 0.5 h at 5 00C. Figure 6 9 A shows a HAADF STEM image of the Ni only gate HEMT indicating gate metal penetration, and Figure 6 9 B shows a HAADF STEM image of the Au only gate HEMT indicating no gate metal penetration. These results indicate that Au is either less likely to penetrate the dis locations or Au is less reactive with the dislocation cores. This is furt her evidence suggesting that once the Au concentration in

PAGE 133

133 the gate metal penetration sufficiently increases that the metal penetration either st ops or is significantly hindered Thus, i n order to avoid these interdiffusion effects on saturating gate metal dislocation penetration the anneal times were limited to short durations of 0.5 h. These times helped to avoid slowing the metal penetration rate due to increased interdiffusion which allowed a more accurate measurement of the metal penetration rate and subsequently the calculated activation energy for metal penetration. Presented in Fig ure 6 10 A F are HAADF STEM images showing the metal penetration into the dislocation at no anneal, 300 C 350 C 400 C 450 C and 500 C at 0.5 h respectively. It can be observed comparing Figure 6 10 B, a 300C anneal for 0.5 h, and Figure 6 10 F, a 500C anneal for 0.5 h, that for the same time the amount of in terdiffusion is much greater in the 500C sample. This is indicated by the brighter contrast of the origina l Ni layer which is now the same contra st as the Au layer in Figure 6 10 F. Due to the nature of the contrast mechanism in HAADF STEM, this increase in brightness in the layer implies that the average Z has increased. This indicates that more Au is now i n the original Ni layer and that the amount of interdiffusion is larger from the high temperature anneal. Add itionally, from Figure 6 10 the rate of penetration of the gate metal into the dislocations was calculated by m easuring the penetration distan ce of the metal and divi di ng the penetration distance by the anneal time, 0.5 h. From Figure 6 10 B F, it is clear that with increasing anneal temperature penetration depth increased. Making the assumption that the metal penetration is a diffusion base d process, then using the Arrhenius equation to plot the logarithm of the penetration rate against the anneal

PAGE 134

134 temperature would yield a linear fit whose slope is the activation energy. Indeed this is the case w hen these metal penetration rates are plotted on a semi log graph as shown in Fig ure 6 11 ; the data appears approximately linear. Upon fitting a curve to the plot of the data, the activation energy for metal penetration was extracted ~0.31 eV It is unsurprising that this calculated activation ener gy is larger than in the pipe diffusion case presented by Pearton et al of 0.23 eV. The activation energy by Pearton et al. is only for O diffusion down the dislocation core from a SiO 2 cap. Additionally, O reacts with GaN at higher temperatures used in their anneal conditions from 500C to 900C. This could result in a lower effective activation energy for O diffusion as the O would be more likely to react with the extra free surfaces in the dislocation core. However, in this work metallic Ni is penetr ating the dislocation cores. While Ni can react and form intermetallics with Ga and Al, measurements of the lattice parameter using HAADF STEM indicate that Ni is only penetrating into the AlGaN and not reacting with the epilayers. Additionally, the Ni p enetration spreads laterally as well while from the Pearton et al. work it is unclear if the O has any lateral spread or if it is only confined to the dislocation core by pipe diffusion. Due to this reaction less penetration, change in layering from GaN t o AlGaN, and lateral expansion of the Ni penetration, the activation energy sho uld be larger in this work compared to previously studies Further work on the possible impact of impurity diffusion on AlGaN/GaN HEMTs was completed in collaboration by Tapaj na et al. and Kuball et al. Together they used trapping measurements and a diffusion based model to explain the increase in trapping as a result of increased impurity diffusion down dislocations. From their model, they calculated an activation energy of 0.26 eV for impurity diffusion down dislocations.

PAGE 135

135 Again, the activation energy for the work presented here is larger but still within the same order of magnitude. It is also important to note that no physical characterization was completed by Tapajna et al. and Kuball et al.; meaning their derived activation energy is based merely on fitting a curve to a model and not on physical proof of impurity diffusion. Thus, it seems reasonable that the 0.31 eV activation energy presented here is on the correct sca le and larger than previously reported values. 6 .3.3 Iso thermal Anneals In the calculation of the activation energy of the metal penetration, it was assumed that the process was diffusion based due to the set up of the experiment and the linear fit of the data. However, in order to verify the Arrhenius behavior of the metal penetration, iso thermal anneals of varying duration were completed at 300C. These p ost iso thermally annealed HEMTs were subjected to FIB preparation of XTEM samples and HAADF STEM imaging was used to image the progressive metal penetration as anneal time increased in the samples. Fig ure 6 12 A D shows the high magnification HAADF STEM images of the 0.5 h 4 h 8 h and 24 h anneals at 300C. It is clear from these images that t he metal penetration depth is increasing as the anneal duration is increased. Next, the p enetration depth against time was plotted as shown in Fig ure 6 13 From Figure 6 13 it appears that the data follow approximately a ( D T) 1/2 fit as shown by the reg ression equation on the plot The power of the regression equation shows a fit of ~0.59 which is close to 0.5, a (DT) 1/2 Due to this closeness of fit, it is acceptable to conclude that the metal penetration is diffusion limited and that the assumptions used to ca lculate the activation energy are accurate, too.

PAGE 136

136 6.3.4 Effect on I G V T I D and Contact Resistance In additional to the structural imaging, pre annealed and post annealed devices were als o characterized electrically in order to develop relations hips between the change in structural properties (metal penetration and interdiffusion) to electrical changes. First an iso chronal anneal study was completed with HEMTs stressed for 0.5 h at 300C, 400 C, 450C, 500C, and 600C. Before and after the a pplied thermal I G G m V T family of I D curves, contact resistance (by TLM) and V FB (by C V) were measured. Representative I G curves for the annealed HEMTs at 300C, 400C, 450C, 500C, and 600C are presented on a combined plot in Fi gure 6 14 and separately in Figure 6 15 through Figure 6 19 respectively From the combined I G data plot shown in Figure 6 14, it can be seen that under positive V GS the I G is always greater than the unstressed value when V GS > 0.4 V except fo r the 300C thermal stress However, the I G behavior varies when V GS is negative depending upon the stress condition For the 300C anneal shown in Figure 6 15 I G is always lower than the original value indicating that leakage is reduced for all value s Additionally, t he I G from 400C anneal is shown in Figure 6 16 and is always below the unstressed I G These results are unsurprising as the interdiffusion of the Ni Au < 400C is relatively small compared to higher temperatures, and the metal penetr ation is less than 3 nm deep into the AlGaN so not many structural changes have occurred. Additionally, it has previously been reported by Singhal et al. that gate anneals can reduce the I D SS which could be similar to the mechanism taking place here (as t his thermal stress is a gate anneal) 47 48 However, for t he 450C and 500C anneals shown in Figure 6 17 and 6 18, respectively, both cause similar trends in the I G which varies widely. For the first few volts of negative V GS the I G is below the

PAGE 137

137 unstr essed value. However, below V GS = 3 V, the I G is always greater than the unstressed value. Lastly, a s can be seen in Figure 6 1 4 and Figure 6 19 the I G does not always become greater than the initial I G until the 600C anneal. This could be due to interdiffusion of the Ni Au in the gate contact being more complete or the physical degradation of the gate itself due to delamination shown earlier in Figure 6 6 C. V GS = 0 and gener ally speaking when V GS > 1.1 V the V T Thus, it can be seen from Figure 6 14, that as long as the thermal anneal is < 500C for 0.5 h, the I G is decreased when 3 V < V DS < 0.4 V. Therefore, if the shift in other electrical parameters is acceptable, these stress conditions could be used to improve device performance with respect to I G In addition to the I G changes, the V T I D change due to thermal stressing as shown in Table 6 1. Here, the V T shift is the relative change in V T between the stressed and unstressed HEMTs. For example in the 400C anneal, the V T = 0.96 V before stress and V T = 1.22V after stress which resulted in the reported shift of 0.26 V. This shift in V T can be seen in the I D V GS plot in Fi gure 6 20 for the 400C anneal. For the thermal anneals it is shown that at 300C there is almost no V T shift; however by 400C the V T shift is large but decreases in magnitude as the anneal temperature is increased to 500C. For the normalized I D ch ange, the measured I D during saturation at V GS = 0 V after stress was divided by the measured I D during saturation at V GS = 0 V before stress. Thus, if the I D is larger post stress, the normalized I D > 1, and if the I D is smaller post stress, the normaliz ed I D < 1. This change in can be seen in Figure 6 21 for the 400C annealed sample for 0.5 h Additionally, all stress temperature > 300C resulted in a

PAGE 138

138 large r I D except the 450C annealed sample. Furthermo re, as the temperature increased the change in I D S decreases. This is mostly likely due to the shift in V T discussed earlier and not the change in V T shifts more positive with temperature, the gates will be more depleted at V GS = 0 V which will lower the I D For change is e when thermal stress was < 500C and decreased when > 500C. This might explain the outlier at 450 I D change in I D Following these iso chronal anneals, iso thermal anneals at 300C were completed up to 400 h. Again, 300C was chosen to lessen the effects of interdiffusion on the electrical results and to isolate the effects of deeper metal penetration. The normalized I G for t hese HEMTs is presented in Figure 6 22. It is shown that the normalized I G is always lower than the initial measurement except when the gate is under forward bias > 0.7 V. Additionally, the V T slowly shifts more positive as the thermal stressing is con tinued as shown in Figure 6 23. Furthermore it is shown in Figure 6 24 that the normalized I D measured at V GS = 0 V is always lower than the initial measurement. These measurements all run in contrast to previously reported work by Marcon et al. abo ut the change in I G V T and I D 8 This could be indicative that the metal penetrat ion plays a larger role in this degradation compared to previous work, or that the gate metal interdiffusion reported in the previous work was not the true culprit of the deg radation. 6.3.5 Simulation of Metal Penetration Effect on V T and I D In order to b etter understand the effect the gate metal penetration had on the V T and I D alone without the effects from interdiffusion a simulation was performed using

PAGE 139

139 the Florida Object Oriented Device Simulator. 100 101 Here, the gate was modified so that a small piece penetrated the AlGaN epilayer. Due to system restrictions, the penetration was approximated to be rectangular instead of triangul ar as would be observed in a XTEM image due to the defects hexagonal pyramid shape. Also, the 3D features of the gate were normalized to the 2D simulation grid. The derivation of these 2D features from the 3D HEMTs is shown in Appendix B. Lastly, m easur ed electrical data was used to calibrate the simulated results to actual device performance An image of the simulated gate design for no gate metal penetration, 5 nm wide x 2.4 nm of penetration, and 5 nm wide x 3.5 nm of penetration are shown in Figure 6 25 A C, respectively. Figure 6 25 B corresponds to the penetration from the 500C for 0.5 h anneal and Figure 6 25 C corresponds to the penetration from the 300 C for 24 h. For these penetration depths, the simulated shifts in the V T are 0.015 V and 0.042 V, respectively. Since the 500C for 0.5 h had th e deepest penetration in the anneals for 0.5 h and simulation shows only a 0.015 V change in V T it can be concluded that the negative shift in V T from the 0.5 h anneals is not from the metal p enetration. However, this shift could be due to the Ni Au interdiffusion, a change in the charge distribution in the interfacial layer, a change in the surface or buffer states due to annealing, or a possible change in the Fermi level pinning of the gate contact. T hese results indicate that the gate metal penetration causes very little change in the V T and is not the primary factor for th e shift. Additionally, the I D V D family of curves were simulated for the no gate metal penetration, 5 nm wide x 2.4 n m of penetration, and 5 nm wide x 3.5 nm of penetration. It was found that for 2.4 nm of penetration the I D decreased by 1.6 % at V GS = 0. For

PAGE 140

140 3.5 nm of penetration the I D decreased by 4.4 % at V GS = 0. However, from the experimental data in Table 6 1, it is known that the I D actually increased from annealing. Thus, it appears that the metal penetration is not responsible for this effect either. 6.3.6 Capacitance Voltage Results Because the simulations could not explain the shit in V T for the HE MTs C V measurements were performed to measure the change in V T and effective E F From the C V measu rements, the change in flat band voltage, V FB was measured. The V FB is defined as: V FB MS Q f C 1 1 0 t (6 1 ) MS is the difference in work function between the metal and semiconductor, Q f is charge in the interface and the layer, C is the capacitance of the layer, is the permittivity of the layer, t is the thickness of the layer, and is the cha rge density within the layer. However, MS can vary due to Fermi level pinning ; therefore the specific change in MS from these measurements cannot be ascertained as they may vary due to pinning Additionally, the change in trap charges in the interface and the layer cannot be separated from the possible change in Fermi level pinning or the change in the metal work function due to Ni Au interdiffusion. Therefore, the measured value of V FB from the C V could be written as: V FB = FB effective E F (6 2 ) Thus, V T and V FB are related by: V T ~ V FB + effective E F T (simulation) ( 6 3 ) C V measurements were completed before and after thermal anneals at 300 C 400 C 450 C 500 C and 600 C at 1 MH z The shift in the V FB was calculated for each

PAGE 141

141 anneal and is listed in Table 6 1 Representative C V curves before and after the 400C anneal for 0.5 h are shown in Figure 6 26. Interestingly, the shifts in the V FB match the trend shown by the shift in the V T between 400C and 500C. Both show a large negative spike in the voltages at 400C and continue to shift more positive towards zero net change as the temperature increases to 500C. This indicates that the shift in V FB and t herefore effective E F is partly responsible for the shift in V T as the only change in V T shown by the gate metal penetration from the simulations was only slightly positive. Additionally, from Figure 6 26, it can be seen that the before and afte r anne al C V curves are similar in shape and only a horizontal shift has occurred. Normally, when this shift is toward a more negative value, this is due to an increase in the charge of the capacitive layer. This indicates that part of the effective E F shift c ould be due to charge creation from the thermal anneals. Additionally, Au has a larger work function than Ni by ~0.3 V. Thus, as more interdiffsion occurs at higher temperatures the work function should shift larger which would make the V FB and therefore the V T shift more positive. Therefore, the relative positive shift observed in the and from 400C to 500C could be due to increased interdiffusion. However, should large Fermi level pinning occur in these HEMTs, this effect would be mitigated or neglig ible although the increase of Au in the metal penetration could neutralize some of the charge or traps which again would lead to the relative positive shift with increased temperature. Further thermal studies paired with C V and additionally electrical me asurement should be completed in the future to separate the differe nt factors E F including but not limited to isolating : the effect of the change in metal work function from the Ni Au

PAGE 142

142 interdiffusion, the effect of the change in charg e and trapping in the layers, the effect of the change in surface state charge, and effect of the change in Fermi level pinning. 6 .4 Summary The application of thermal anneals to HEMTs has produced many meaningful results to improve the reliability of AlG aN/GaN HEMTs. One of the most meaningful changes is the degradation of the gate structure. For example, at temperatures greater than 550C the gate delaminates from the AlGaN epilayer. This is an extreme case of failure that degrades the electrical pro perties of the HEMT and should be avoided Another structural change is the interdiffusion of the Ni and Au in the gate stack. This effect could influence the work function of the metal contact or passivate trapped charges in the material which would aff ect the electrical proerties. Furthermore, thermal anneals cause d the gate metal to penetrate the AlGaN epilayer. This metal penetration causes many electrical changes in the H EMTs. It was found upon annealing that the V T shifted negatively for tempera tures > 400C, and for long thermal anneals at 300C up to 400 h the V T shifted positively. Additionally, I DS i ncreased for anneals > 400C but decreased for long term anneals at 300C. Using simulations of gate metal penetration, it was found that metal penetration only play ed a minor role in V T shift, a maximum of +0.04 V. Additionally, the s imulations showed that only slight decreases in I DS occurred, up to 4 .4 %. Thus, C V measurements were F V measurements showed that the V T and V FB /effective E F shift followed the same trends and were on the same order of magnitude. This indicates that the V T shift is most likely due to the change in effective E F from the annealed HEMTs.

PAGE 143

143 Table 6 1 Electrical data from 0.5 h anneals on Si(111) AlGaN/GaN HEMTs Temperature ( C) 300 400 450 500 V T shift (V) 0.03 0.26 0.16 0.10 I D / I D (0) 0.98 1.33 0.79 1.19 3.8 2.0 4.2 0.29 V FB shift (V) 0.06 0.15 0.13 0.12

PAGE 144

144 Figure 6 1. Low magnification XTEM images of the gate region of AFRL AlGaN/GaN HEMT s on SiC showing the distinct Ni/Au gate metal stack layers a nd AlGaN and GaN epilayers A) Unannealed HEMT. B ) HEMT ann ealed at 750C for 0.5 h C ) Different H EMT annealed at 750C for 0.5 h. Figure 6 2. Low magnification HAADF STEM images of th e gate region of AFRL AlGaN/GaN HEMTs on SiC showing the distinct Ni/Au gate metal stack layers and AlGaN and GaN epilayer s A ) U nannealed HEMT. B ) HEMT annealed at 750C for 0.5 h C ) Different HEMT annealed at 750C for 0.5 h. Both annealed HEMTs sho w a modification of the Ni layer of the gate but no change to the AlGaN

PAGE 145

145 Figure 6 3. PSEM images of the top of the AlGaN layer A ) U nannealed sub m gate le ngth HEMT. B ) A nnealed sub m HEMT showing no physical changes to the AlGaN layer C ) A nneal ed sub m HEMT showing degradation to the AlGaN epilayer.

PAGE 146

146 Figure 6 4. PSEM images of the top of the AlGaN layer of 1 m gate length AFRL AlGaN/GaN HEMTs on SiC. A ) U nannealed B ) A nnealed at 450C for 0.5 h showing no physic al changes to the AlGaN layer. C ) A nnealed at 650C for 0.5 h showing slight pitting under the gate indicated by red arrows D ) A nnealed at 750C for 0.5 h showing large, copious pitting only under the gate

PAGE 147

147 Figure 6 5. High magnification PSEM image of the top of the AlGaN layer of an AFRL AlGaN/GaN HEMT on SiC showing the hexagonal symmetry of the pits indicated by white arrows.

PAGE 148

148 Figure 6 6 HAADF STEM images of the gate region and epilayers of Ni gate AFRL AlGaN/GaN HEMTs on Si(111) showi ng the gate metal penetration in dislocations A ) L ow magnification image after a 500 C anneal for 0.5 h. B ) H igh magnification image from the box in part A clearly showing the gate metal penetration. C ) L ow magnification image showing the partial delamination of the gate from the A lGaN epilayer after a 600C anneal for 0.5 h

PAGE 149

149 Figure 6 7 High magnification HAADF STEM images of the gate /AlGaN interface of AFRL HEMTs on Si(111) showing the gate metal penetration depth into the AlGaN epi layer after a 500C anneal A) 0.5 h B ) 6 h. The penetration depth of the images is within 1 nm.

PAGE 150

150 Figure 6 8. XTEM image of gate metal penetration and EDS chemical composition analysis. A) HAADF STEM image of the gate metal penetration at 500C for 0.5 h indicating the area of the EDS line scan. B) EDS line scan of the gate metal penetration showing the composition in the AlGaN and metal which is mostly Au

PAGE 151

151 Figure 6 9. HAADF STEM images of only single metal coated AlGaN/GaN epilayers after a 500 C anneal for 0.5 h A) Ni. B) Au.

PAGE 152

152 Fi gure 6 10 High magnification HAADF STEM images of the gate/AlGaN interface of HEMTs on S i(111) after a 0.5 h anneal A) No anneal exposure. B) 300C. C) 350C. D) 40 0C. E) 450C. F ) 500C The dashed lines delineate the outline of the increasin g metal penetration into the dislocation and increased Ni Au interdiffusion is observed with temperature

PAGE 153

153 Figure 6 11 Plot of the rate of metal penetration vs. temperature for 0.5 h iso chronal anneals for HEMTs on Si(111) from 300C to 500C. An ac tivation energy for this process is linearly extracted from this plot indicating a value of ~0.31 eV.

PAGE 154

154 Figure 6 12 High magnification HAADF STEM images showing the gate metal penetration after a 300C anneal A ) 0.5 h B ) 4 h C ) 8 h D ) 24 h

PAGE 155

155 Figure 6 1 3 Plot of the gate metal penetration distance vs. duration of anneal at 300C. The regression equation shows that the fit is near square root DT indicating that the penetration is diffusion limited.

PAGE 156

156 Figure 6 14 Normalized I G plot showin g anneals at 300C, 400C, 450C, 500C, and 600 C for 0.5 h In general, t he I G increases with temperature.

PAGE 157

157 Figure 6 15. Change in I G after a 0.5 h anneal at 300C indicating that the I G is always lower than the unstressed condition.

PAGE 158

158 Figure 6 1 6. Change in I G after a 0.5 h anneal at 4 00 C indicating that the I G is always smaller than the unstressed condition when V GS < 0.7 V

PAGE 159

159 Figure 6 17. Change in I G after a 0.5 h anneal at 45 0 C. Figure 6 18. Change in I G after a 0.5 h anneal at 5 00 C.

PAGE 160

160 Figure 6 19. Change in I G after a 0.5 h anneal at 600 C indicating that the I G is always larger than the unstressed condition Figure 6 20. Representative V T shift in an AlGaN/GaN HEMT annealed at 400C for 0.5 h.

PAGE 161

161 Figure 6 21. Normalized I D from the 400 C anneal for 0.5 h indicating that the I D has increased from the stress.

PAGE 162

162 Figure 6 22. Normalized I G for HEMTs exposed to a 300C anneal for up to 400 h. The figure indicates that all HEMTs exhibited lower I G below V GS = 0.7 V.

PAGE 163

163 F igure 6 23. Shift in V T over time for HEMTs exposed to a 300C anneal up to 400 h.

PAGE 164

164 Figure 6 24 Normalized I D for HEMTs exposed to a 300C anneal for up to 400 h. The figure indicates that I D decreased with time.

PAGE 165

165 Figure 6 25 G ate structure s us ed in s imulation s to determine the effect of gate metal penetration on HEMTs A) No penetration. B) Gate metal penetration 5 nm wide and 2.4 nm deep representing the 500C for 0.5 h anneal. C) Gate metal penetration 5 nm wide and 3.5 nm deep representin g the 300C for 24 h anneal.

PAGE 166

166 Figure 6 26. Representative C V curves from an AFRL HEMT on Si(111) before and after a 400C anneal for 0.5 h.

PAGE 167

167 CHAPTER 7 CONCLUSIONS AND FUTURE WORK The body of this presented work as has identified many reliability an d failure mechanisms in AlGaN/GaN HEMTs. Additionally, how these defects affect the device performance has been outline. These defects have been shown to be as grown in the devices, caused by electrical stress, or caused by thermal stress. Interfacial l ayers were characterized using TEM and APT. It was revealed that they consisted of dual NiO x and AlO x layers. Furthermore, in different device structures with a similar Ni gate, a native oxide later was chemically identified using EELS analysis between t he gate and AlGaN epilayer. Unfortunately, in these devices, APT analysis was not possible, but the HR TEM still ultimately revealed the O present in the interface. By understanding the composition of this interfacial layer, it can now be removed from th e HEMT fabrication process which will improve performance and reliability. I t was shown that applying large electric fields can cause the Ni from the gate to react with the AlGaN epilayer. Additionally, it is noted that the reaction was an oxide and tha t N was deficient in the defect compared to the unstressed state. This implies that the Ni and AlGaN mixed into an oxide. Furthermore, the shape of the defects matched the shape of the electric field s between the 5 6 MV/cm contours Therefore, care sh ould be taken to design devices so that the magnitude of the field is mitigated or spread out so that i t does not become too large to control the shape of a Ni/AlGaN reaction defect should one occur Additionally, finding these large scale under gate rea ction defects is challenging TEM and APT both rely on < 100 nm of sample material Therefor e, identifying defects by serially sectioning the samples using these techniques would prove fruitless. However, analyzing the electrical data can show if

PAGE 168

168 defect s may be present. It was shown that large, rapid initial decreases to the normalized I D max of stressed HEMTs corresponded to the prese nce of these reaction defects under the gate in HEMTs. Additionally, it was shown that when the I D max decreased slowly that these defects were never found. Furthermore, all HEMTs experience d an I G jump from the electrical stressing if carried out above the critical voltage. When the electrical stress was stopped before, during, and after the I G jump and the normalized I D max These results indicate that monitoring the normalized I D max a better measure for detecting these reaction based under gate defects. Lastly, it was shown that gate metal penetrated the thr eading dislocation in the HEMTs. This pe netration negatively affects the device electrical performance although a combination of thermal stress followed by electrical stress was neve r performed. It was found upon annealing th at the V T shifted negatively for temperatures > 400C, and for long thermal anneals at 300C up to 400 h the V T shifted positively. Using simulations of the gate metal penetration, it was found that penetration only play a minor role in the V T shift, a ma ximum of 0.04 V. Thus, C V measurements were performed to F V measurements showed that the V T and V FB /effective E F shift followed the same trends. This indicates that the V T shift is most likely due to the chan ge in effective E F from the annealed HEMTs. While t here are many reliability issues presented here, there are still many issues that remain wi th these devices. Future work sh ould explore the effect of changing the device materials or structure on the form ation of these defects and therefore their influence on device performance. One of the first studies that should be undertaken is

PAGE 169

169 replacing the Ni gate metal. It is clear from the Ni reactions with the AlGaN in Chapter 5 and the Ni penetration into the d islocations in Chapter 6 that Ni is too reactive with AlGaN to be stable for long term reliability. Therefore, it should be replaced with a metal less reactive with O or AlGaN. While passivation plays an important role in decrea sing surface traps, it also functions as a diffusion barrier to the ambient and contains contamination. Studies should be undertaken to test different chemistries of passivation from oxide and ni tride based to more exotic ones. For example C hapter 5 showed that the Ni and AlGaN c an react with O and form a defect. Therefore, it is important to find out if an oxide based passivation like ALD Al 2 O 3 reacts with the gate metal or AlGaN epilayer. Additionally its effectiveness as a diffusion barrier needs to be ascertained While pa ssivation that does not react with the AlGaN is great for reliability, i f it does not stop ambient from reaching the surface then reactions could still occur. Additionally, due to the inverse piezoelectric effect the amount of stress each type of passiva tion induces should be studied. Further studies should focus on reducing the dislocation density in AlGaN/GaN HEMTs In Chapter 6, it was shown that gate metal penetrat es dislocations. Therefore, more dislocations present in the device will result in mor e metal penetration into the AlGaN epilayer and the more negative effect on electrical properties and reliability these defects will produce By decreasing the dislocations, less metal penetration will occur and the HEMTs will become more reliable Furt hermore, different types of gate metal stacks should be examine d to observe if different metals penetrate dislocations at different rates as Ni is shown to penetrate the dislocations much more readily than Au.

PAGE 170

170 An extension of this work would be the chang e the passivation layer chemistry between the gate and AlGaN. Different interfacial chemistries may increase or decrease the rate of penetration due to the diffusivity or reactivity of the different metals with the layer.

PAGE 171

171 APPENDIX A FLOOPS SCRIPT S FOR G ATE STRUCTURES In this appendix, the FLOOPS script s for eac h gate structure in this work are included. These gat e structures include the T gate with 100 nm length on SiC, the 1 m gate length on SiC, 1 m gate length on Si(111), and an example of metal pe netration with 5 nm width and 2.4 nm depth from a 1 m gate length on Si(111) The bolded text indicates the type of structure and is not part of the code T gate, 100 nm AFRL HEMT on SiC proc Struct2D {} { ;# T gate, Lg=100 nm, SiC #Structure definition line x loc= 0.291 spac=0.01 tag=MetTop line x loc= 0.087 spac=0.01 tag=Oxtop line x loc=0.0 spac=0.001 tag=AlGaNTop line x loc=0.014 spac=0.001 tag=AlGaNBottom line x loc=1.8 spac=0.2 tag=GaNBottom line x loc=2.0 spac=0 .02 tag=AlNBottom line x loc=3.0 spac=1.8 tag=BBottom line y loc= 2.0 spac=0.1 tag=Left line y loc= 1.25 spac=0.1 line y loc= 0.2155 spac=0.1 tag=Oxleft

PAGE 172

172 line y loc= 0.0475 spac=0.001 tag=Oxgs line y loc=0.0 spac=0.005 line y l oc=0.0475 spac=0.001 tag=Oxgd line y loc=0.2155 spac=0.1 tag=Oxright line y loc=1.25 spac=0.1 line y loc=2.0 spac=0.1 tag=Right #Bulk region SiC xlo=AlNBottom xhi=BBottom ylo=Left yhi=Right #thin AlN layer region AlN xlo=G aNBottom xhi=AlNBottom ylo=Left yhi=Right #Buffer region GaN xlo=AlGaNBottom xhi=GaNBottom ylo=Left yhi=Right #AlGaN under gate region AlGaN xlo=AlGaNTop xhi=AlGaNBottom ylo=Left yhi=Right #Oxides for t gate region oxide xlo=Oxto p xhi=AlGaNTop ylo=Left yhi=Oxgs region oxide xlo=Oxtop xhi=AlGaNTop ylo=Oxgd yhi=Right region oxide xlo=MetTop xhi=Oxtop ylo=Left yhi=Oxleft

PAGE 173

173 region oxide xlo=MetTop xhi=Oxtop ylo=Oxright yhi=Right #metal in the middle region metal xlo =Oxtop xhi=AlGaNTop ylo=Oxgs yhi=Oxgd region metal xlo=MetTop xhi=Oxtop ylo=Oxleft yhi=Oxright #initialize the grid init #Smooth AlGaN #Smooth GaN #Smooth SiC #Contacts contact name=G metal xlo= 0.28 xhi=0.001 ylo= 0.44 yhi=0.44 add depth=1.0 width=1.0 contact name=B SiC xlo=2.9 xhi=7.0 add depth=1.0 width=1.0 contact name=S AlGaN ylo= 3.4 yhi= 1.99 xlo= 1.0 xhi=0.0145 add depth=1.0 width=1.0 contact name=D AlGaN ylo=1.99 yhi=3.4 xlo= 1.0 xhi=0.0145 add depth= 1.0 width=1.0 }

PAGE 174

174 1 m AFRL HEMT on SiC proc Struct2D {} { ;# Lg=1 um, SiC line x loc= 0.3 spac=0.01 tag=MetTop line x loc= 0.1 spac=0.01 tag=Oxtop line x loc=0.0 spac=0.015 tag=AlGaNTop line x loc=0.015 spac=0.005 tag=AlGaNBottom line x loc=1.8 spac=0.55 tag=GaNBottom line x loc=2.0 spac=0.02 tag=AlNBottom line x loc=3.0 spac=1.5 tag=BBottom line y loc= 2.0 spac=0.05 tag=Left line y loc= 1.25 spac=0.1 line y loc= 0.75 spac=0.1 tag=Oxleft line y loc= 0.5 s pac=0.01 tag=Oxgs line y loc=0.0 spac=0.01 line y loc=0.5 spac=0.01 tag=Oxgd line y loc=0.75 spac=0.1 tag=Oxright line y loc=1.25 spac=0.1 line y loc=2.0 spac=0.05 tag=Right #Bulk region SiC xlo=AlNBottom xhi=BBottom ylo=Left yhi=Right

PAGE 175

175 #thin AlN layer region AlN xlo=GaNBottom xhi=AlNBottom ylo=Left yhi=Right #Buffer region GaN xlo=AlGaNBottom xhi=GaNBottom ylo=Left yhi=Right #AlGaN under gate region AlGaN xlo=AlGaNTop xhi=AlGaNBottom ylo=Left yhi=R ight #Oxides for t gate region oxide xlo=Oxtop xhi=AlGaNTop ylo=Left yhi=Oxgs region oxide xlo=Oxtop xhi=AlGaNTop ylo=Oxgd yhi=Right region oxide xlo=MetTop xhi=Oxtop ylo=Left yhi=Oxleft region oxide xlo=MetTop xhi=Oxtop ylo=Oxright yh i=Right #metal in the middle region metal xlo=Oxtop xhi=AlGaNTop ylo=Oxgs yhi=Oxgd region metal xlo=MetTop xhi=Oxtop ylo=Oxleft yhi=Oxright init #Smooth AlGaN #Smooth GaN

PAGE 176

176 #Smooth SiC #Contacts contact name=G metal xl o= 0.29 xhi=0.001 ylo= 0.76 yhi=0.76 add depth=300.0 width=1.0 ;#for 1um length #contact name=G metal xlo= 0.19 xhi=0.001 ylo= 0.75 yhi=0.75 add depth=300.0 width=1.0 contact name=B SiC xlo=2.4 xhi=9.0 add depth=1.0 width=300.0 contact name=S A lGaN ylo= 3.4 yhi= 1.99 xlo= 1.0 xhi=0.01 add depth=300.0 width=1.0 contact name=D AlGaN ylo=1.99 yhi=3.4 xlo= 1.0 xhi=0.01 add depth=300.0 width=1.0 } 1 m AFRL HEMT on Si(111) proc Struct2D {} { ;# Lg=1.1 um, Si substrate #Structure defi nition: no defect line x loc= 0.30 spac=0.01 tag=MetTop line x loc= 0.01 spac=0.01 line x loc=0.0 spac=0.001 tag=AlGaNTop line x loc=0.014 spac=0.001 tag=AlGaNBottom

PAGE 177

177 line x loc=1.0 spac=0.2 line x loc=2.264 spac=0.2 tag=GaNBot tom line x loc=3.0 spac=0.2 tag=BBottom line y loc= 1.7 spac=0.1 tag=Left line y loc= 1.25 spac=0.1 line y loc= 0.55 spac=0.001 tag=G l line y loc= 0.2 spac=0.1 line y loc=0.0 spac=0.001 line y loc=0.2 spac=0.1 line y loc= 0.55 spac=0.001 tag=G r line y loc=1.25 spac=0.1 line y loc=1.7 spac=0.1 tag=Right #Bulk region Silicon xlo=GaNBottom xhi=BBottom ylo=Left yhi=Right #Buffer region GaN xlo=AlGaNBottom xhi=GaNBottom ylo=Left yhi=Right #AlGaN under gate region AlGaN xlo=AlGaNTop xhi=AlGaNBottom ylo=Left yhi=Right

PAGE 178

178 #passivation region oxide xlo=MetTop xhi=AlGaNTop ylo=Left yhi=G l region oxide xlo=MetTop xhi=AlGaNTop ylo=G r yhi=Right #metal region metal xlo=MetTop xhi= AlGaNTop ylo=G l yhi=G r #initialize the grid init #Contacts contact name=G metal xlo= 0.28 xhi=0.001 ylo= 0.57 yhi=0.57 add depth=300.0 width=1.0 contact name=B Silicon xlo=2.9 xhi=7.0 add depth=300.0 width=1.0 contact name= S AlGaN ylo= 3.4 yhi= 1.7 xlo= 1.0 xhi=0.013 add depth=300.0 width=1.0 contact name=D AlGaN ylo=1.7 yhi=3.4 xlo= 1.0 xhi=0.013 add depth=300.0 width=1.0 } 1 m AFRL HEMT on Si(111) with gate metal penetration 2.4 nm deep and 5 nm wide into AlGaN epila yer proc Struct2D_d1 {} { ;#Lg=1.1 um, Si substrate

PAGE 179

179 #Structure definition:5 nm wide, 2.4 deep defect into AlGaN line x loc= 0.30 spac=0.01 tag=MetTop line x loc= 0.01 spac=0.01 line x loc=0.0 spac=0.001 tag=AlGaNTop line x loc=0.0024 spac=0.0007 tag=defect line x loc=0.014 spac=0.001 tag=AlGaNBottom line x loc=1.0 spac=0.2 line x loc=2.264 spac=0.2 tag=GaNBottom line x loc=3.0 spac=0.2 tag=BBottom line y loc= 1.7 spac=0.1 tag=Left line y loc= 1.25 spac=0.1 line y loc= 0.55 spac=0.001 tag=G l line y loc= 0.2 spac=0.1 line y loc= 0.0025 spac=0.0005 tag=defect l line y loc=0.0 spac=0.0005 line y loc=0.0025 spac=0.0005 tag=defect r line y loc=0.2 spac=0.1 line y loc=0.55 spa c=0.001 tag=G r line y loc=1.25 spac=0.1 line y loc=1.7 spac=0.1 tag=Right

PAGE 180

180 #Bulk region Silicon xlo=GaNBottom xhi=BBottom ylo=Left yhi=Right #Buffer region GaN xlo=AlGaNBottom xhi=GaNBottom ylo=Left yhi=Right #AlGaN under ga te region AlGaN xlo=AlGaNTop xhi=AlGaNBottom ylo=Left yhi=defect l region AlGaN xlo=AlGaNTop xhi=AlGaNBottom ylo=defect r yhi=Right region AlGaN xlo=defect xhi=AlGaNBottom ylo=defect l yhi=defect r #passivation region oxide xlo=MetTop xhi=AlGaNTop ylo=Left yhi=G l region oxide xlo=MetTop xhi=AlGaNTop ylo=G r yhi=Right #metal region metal xlo=MetTop xhi=AlGaNTop ylo=G l yhi=G r region metal xlo=AlGaNTop xhi=AlGaNBottom ylo=defect l yhi=defect r #initialize the gri d init

PAGE 181

181 #Contacts contact name=G metal xlo= 0.28 xhi=0.01 ylo= 0.57 yhi=0.57 add depth=300.0 width=1.0 contact name=B Silicon xlo=2.9 xhi=7.0 add depth=300.0 width=1.0 contact name=S AlGaN ylo= 3.4 yhi= 1.7 xlo= 1.0 xhi=0.013 add d epth=300.0 width=1.0 contact name=D AlGaN ylo=1.7 yhi=3.4 xlo= 1.0 xhi=0.013 add depth=300.0 width=1.0 }

PAGE 182

182 APPENDIX B DERIVATION OF GATE METAL PENETRATION DEFECT SIZE FOR SIMULATION FLOOPS uses a 2D representation of a 3D transistor. Therefore, the 3D transistor must be approximated and projected into 2D. This is taken into account in two ways when determining the size of the defects for the gate metal penetration simulation First the average width of the defect must be calculated. Second, the avera ge depth of the defect must be calculated. For the first condition, the average width of the defect is found by finding the total area under the gate of the defects simulation gate length. The total area of the defects can be found by: A total = A defect A gate (B 1) where A defect is the area of a regular hexagon which approximates the shape of the defect and is known to be 3 3/2 a 2 2 1 where a is the length of a side of the hexagon (nm) 2 ). The area of the gate is known to be 165 m 2 from the gate length of 1.1 m and gate width of 150 m. If approximately some amount, x, where x is on the order of 10 9 cm 2 then it can be shown that: = ( A defect ) ( A gate ) ( A gate 1 ) ( B 2) = (A defect ) ( ) (B 3) = (3 3/2 a 2 2 1 ) (x 10 9 ) (B 4 ) = (3 3/2 a 2 2 1 ) (x 10 9 ) ( 10 1 4 ) (B 5 ) = (3 3/2 a 2 2 1 ) (x 10 5 ) (B 6 )

PAGE 183

183 is the proportion of the gate length o f the simulation defect. If a = 13 nm and x = 1 (therefore a density of 10 9 cm 2 width of the simulation defect would be: G (B 7) nm) (0.0044) (B 7) The average depth for the gate metal penetration can also be similarly calculated given by: defect A defect 1 (B 8) where V defect is the volume of the defect. Again, the defect is approximately a regular hexagonal pyramid based upon SEM and TEM analysis. Therefore, the V defect can be calculated by findin g the volume of a regular hexagonal pyramid. A regular hexagonal pyramid can be divided into 12 equal parts due to symmetry and the volume for each part can be found by: V part = 0 a/2 r x r a/2 (h 2 h y a 1 r 1 ) dy dx (B 9) where r = 3 1/2 a is the length of a side of the regular hexagon al base, and h is the height of the pyramid (or in this case the depth of the metal penetration) This results in V defect = 8 1 3 1/2 h a 2 Now the V defect can be found by: V defect = 12 (V part ) (B 10) where multiplying V part by 12 was from the original symmetry of the hexagonal pyramid to simplify the calculations for the v olume. T his yields V defect = 3 1/2 2 1 h a 2 Now, solving for the average penetration depth, anneal at 500 C for

PAGE 184

184 24 h, the penetration depth was ~11 nm used in the simulations.

PAGE 185

185 LIST OF REFERENCES 1 M. Feng, S. Shyh Chiang, D. C. Caruth and J. J. Huang, Proc. IEEE 92 354 (2004). 2 J. W. Johnson, E. L. Piner, A. Vescan, R. Therrien, P. Rajagopal, J. C. Roberts, J. D. Brown, S. Singhal and K. J. Linthicum, IEEE Electron Device Lett. 25 459 (2004). 3 W. Nagy, J. Brown, R. Borges and S. Singhal, IEEE Trans. Microw. Theory Tech. 51 6 60 (2003). 4 Nidhi, S. Dasgupta, S. Keller, J. S. Speck and U. K. Mishra, IEEE Electron Device Lett. 32 1683 (2011). 5 S. Tirelli, D. Marti, H. Sun, A. R. Alt, J. F. Carlin, N. Grandjean and C. R. Bolognesi, IEEE Electron Device Lett. 32 1364 (2011). 6 Y Hao, L. Yang, X. Ma, J. Ma, M. Cao, C. Pan, C. Wang and J. Zhang, IEEE Electron Device Lett. 32 626 (2011). 7 D. Maier, M. Alomari, N. Grandjean, J. F. Carlin, M. A. di Forte Poisson, C. Dua, A. Chuvilin, D. Troadec, C. Gaquiere, U. Kaiser, S. L. Delage and E. Kohn, IEEE Trans. Device Mater. Reliab. 10 427 (2010). 8 D. Marcon, X. Kang, J. Viaene, M. Van Hove, P. Srivastava, S. Decoutere, R. Mertens and G. Borghs, Microelectron Reliab. 51 1717 (2011). 9 U. K. Mishra, P. Parikh and Y. F. Wu, Proc. IEEE. 90 1022 (2002). 10 U. K. Mishra, L. Shen, T. E. Kazior and Y. F. Wu, Proc. IEEE. 96 287 (2008). 11 G. Meneghes so, G Verzellesi, F. Danesin, F. Rampazzo, F. Zanon, A. Tazzoli, M. Meneghini, and E. Zanoni, IEEE Trans. Dev. Mater. Reliab. 8 332 (2008) 1 2 U. Chowdhury, J. L. Jimenez, C. Lee, E. Beam, P. Saunier, T. Balistreri, P. Seong Yong, L. Taehun, J. Wang, M. J. Kim, J. Jungwoo and J. A. del Alamo, IEEE Electron Device Lett. 29 1098 (2008). 13 J. Joh and J. A. del Alamo, IEEE Electron Device Lett. 29 287 (2008). 14 J. A. del Alamo and J. Joh, Microelectron. Reliab. 49 1200 (2009). 1 5 S. Y. Park, C. Floresca, U. Chowdhury, J. L. Jimenez, C. Lee, E. Beam, P. Saunier, T. Balistreri and M. J. Kim, Microelectron. Reliab. 49 478 (2009). 1 6 C. Y. Chang, T Anderson, J. Hite, L. Lu, C. F. Lo, B. H. Chu, D. J. Cheney, E. A. Douglas, B. P. Gila, F. Ren, G. D. Via, P. Whiting, R. Holzworth, K. S. Jones, S. Jang and S. J. Pearton, J. Vac. Sci. Technol. B 28 1044 (2010).

PAGE 186

186 1 7 M. R. Holzworth, N. G. Rudawski, S. J. Pearton, K. S. Jones, L. Lu, T. S. Kang, F. Ren and J. W. Johnson, Appl. Phys. Lett. 98 122103 (2011). 18 M. Kuball, M. Tapajna, R. J. T. Simms, M. Faqir and U. K. Mishra, Microelectron. Reliab. 51 195 (2011). 19 E. A. Douglas, C. Y. Chang, D. J. Cheney B. P. Gila, C. F. Lo, L. Lu, R. Holzworth, P. Whiting, K. Jones, G. D. Via, J. Kim, S. Jang, F. Ren and S. J. Pearton, Microelectron. Reliab. 51 207 (2011). 20 E. A. Douglas, C. Y. Chang, B. P. Gila, M. R. Holzworth, K. S. Jones, L. Liu, J. Kim, S. Jang G. D. Via, F. Ren and S. J. Pearton, Microelectron. Reliab. 52 23 (2012). 21 P. G. Whiting, N. G. Rudawski, M. R. Holzworth, S. J Pearton, K. S. Jones, L. Liu, T. S. Kang and F. Ren, Microelectron. Reliab. 52 2542 (2012). 22 F. Bernardini and V. Fioren tini, Phys. Rev. B 56 R10024 (1997 ). 23 A. Sarua, H. Ji, M. Kuball, M. J. Uren, T. Martin, K. P. Hilton, and R. S. Balmer, Appl. Phys. Lett. 88 103502 (2006). 24 B. Jogai, J. D. Albrecht, and E. Pan, J. Appl. Phys. 94 3984 (2003). 25 A. F. M. Anwar, R. T Webster, and K. V. Smith, Appl. Phys. Lett. 88 203510 (2006). 26 F. Bernardini and V. Fiorentini, Appl. Phys. Lett. 80 4145 (2002). 27 R. F. Davis, A. M. Roskowski, E. A. Preble, J. S. Speck, B. Heying, J. A. Freitas, E.R. Glaser, and W.E. Carlos, Pr oc. IEEE. 90 993 (2002). 28 S.C. Binari, J.M. Redwing, G. Kelner, and W. Kruppa, Electron. Lett. 33 242 ( 1997). 29 R. J. Trew, Proc. IEEE. 90 1032 (2002). 30 P. Rajagopal, T. Gehrke, J.C. Roberts, J.D. Brown, T.W. Weeks, E.L. Piner, and K.J. Linthicum Mat. Res. Soc. Symp. Proc. 743 L1.2.1 (2002). 31 Y.C. Chou, D. Leung, I. Smorchkova, M. Wojtowicz, R. Grundbacher, L. Callejo, Q. Kan, R. Lai, P.H. Liu, D. Eng, and A. Oki, Microelectron. Reliab. 44 1033 (2004). 32 H. Morkoc, Handbook of Nitride Semico nductors and Devices, Wiley, New York, 2008, p.483. 33 S.R. Lee, D.D. Koleske, K.C. Cross, J.A. Floro, K.E. Waldrip, A.T. Wise, and S. Mahajan, Appl. Phys. Lett. 85 6164 (2004).

PAGE 187

187 34 C. Lee, L. Witkowski, H. Q. Tserng, P. Saunier, R. Birkhahn, D. Olson, D. O lson, G. Munns, S. Guo, and B. Albert, Electron. Lett. 41 155 (2005). 35 D.W. Gotthold, S.P. Guo, R. Birkhahn, B. Albert, D. Florescu, and B. Peres, J. Electron. Mater. 33 408 (2004). 36 J.L. Jimenez and U. Chowdhury, Proc. IEEE Int. Rel. Phys. Symp. 429 (2008). 37 E. T. Yu, X. Z. Dang, L. S. Yu, D. Qiao, P. M. Asbeck, S. S. Lau, G. J. Sullivan, K. S. Boutros and J. M. Redwing, Appl. Phys. Lett. 73 1880 (1998). 38 H. Zhang and E. T. Yu, J. Appl. Phys. 99 014501 (2006). 39 S. C. Binari, K. Ikossi, J. A. Roussos, W. Kruppa, D. Park, H. B. Dietrich, D. D Koleske, A. E. Wickenden, and R. L. Henry, IEEE Trans. Electron Devices, 48 465 (2001). 40 K. Lee, S. Lee, and K. Yang, MINT MIS 2007. 205 (2007). 41 M. A. Khan, M. S. Shur, Q. C. Chen, and J. N. Kuznia, E lectron Lett. 30 2175 (1994). 42 S. C. Binari, W. Kruppa, H. B. Dietrich, G. Kelner, A. E. Wickenden, and J. A. Freitas, Jr., Solid State Electron. 41 1549 (1997). 43 S. C. Binari, P. B. Klein, and T. E. Kazior, Proc. IEEE, 90 1048 (2002). 44 S. K. Jha, B. H. Leung, C. Surya, H. Schweizer, and H. Pilkhuhn, Proc. COMMAD, 33 (2004). 45 A. Sozza, C. Dua, E. Morvan, B. Grimber, and S. L. Delage, Microelectron Reliab. 45 1617 (2005). 46 S. Yoshida and J. Suzuki, J. Appl. Phys. 84 2940 (1998). 47 S. Singhal, T. Li, A. Chaudhari, A.W. Hanson, R. Therrien, J.W. Johnson, W. Nagy, J. Marquart, P. Rajagopal, J.C. Roberts, E.L. Piner, I.C. Kizilyalli, and K.J. Linthicum, Microelectron. Reliab. 46 1247 (2006). 48 S. Singhal, J.C. Roberts, P. Rajagopal, T. Li, A.W. Hanson, R. Therrien, J.W. Johnson, I.C. Kizilyalli, and K.J. Linthicum, Proc. IEEE Int. Rel. Phys. Symp. 95 (2006). 49 M. Piazza, C. Dua, M. Oualli, E. Morvan, D. Carisetti and F. Wyczisk, Microelectron. Reliab. 49 1222 (2009).

PAGE 188

188 50 B. Luo, F. Ren, R.C. Fit ch, J.K. Gillespie, T. Jenkins, J. Sewell, D. Via, A. Crespo, A.G. Baca, R.D. Briggs, D. Gotthold, R. Birkhahn, B. Peres and S.J. Pearton, Appl. Phys. Lett. 82 3910 (2003). 51 J. Joh and J. A. del Alamo, IEDM Tech. Dig. 415 (2006). 52 J. Joh and J. A. del Alamo, IEDM Tech. Dig. 385 (2007). 53 J. A. del Alamo and J. Joh, Microelectron Reliab. 49 1200 (2009). 54 M. Faqir, G. Verzellesi, A. Chini, F. Fantini, F. Danesin, G. Meneghesso, E. Zanoni, and C. Dua, IEEE Trans. Device Mater. Reliab. 8 240 (2008). 55 D. Marcon, J. Viaene, P. Favia, H. Bender, X. Kang, S. Lenci, S. Stoffels, and S. Decoutere, Microelectron Reliab. 52 2188 (2012). 56 M. A. Mastro, J. R. LaRoche, N. D. Bassim, and C. R. Eddy, Microelectron J. 36 705 (2005). 57 Z. Liliental Weber, Y. Chen, S. Ruvimov, and J. Washburn, Phys. Rev. Lett. 79 2835 (1997). 58 S. J. Pearton, H. Cho, J. R. LaRoche, F. Ren, R. G. Wilson, and J. W. Lee, Appl. Phys. Lett. 75 2939 (1999). 59 M. Tapajna, U. K. Mishra, and M. Kuball, Appl. Phys. Lett. 97 023503 ( 2010). 60 A.M. Conway, M. Chen, P. Hashimoto, P.J. Willadsen, and M. Micovic, CS MANTECH Conference. 99 (2007). 61 S. D. Burnham, R. Bowen, P. Willadsen, H. Bracamontes, P. Hashimoto, M. Hu, D. Wong, M. Chen and M. Microvic, Phys. Status Solidi C. 8 2399 (2011). 62 P. Makaram, J. Joh, J. A. del Alamo, T. Palacios and C. V. Thompson, Appl. Phys. Lett. 96 233509 (2010). 63 L. Li, J. Joh, J. A. Del Alamo and C. V. Thompson, Appl. Phys. Lett. 100 172109 (2012). 64 F. Gao, B. Lu, L. Li, S. Kaun, J. S. Speck, C. V. Thompson and T. Palacios, Appl. Phys. Lett. 99 223506 (2011). 65 D. N. Seidman, Annu. Rev. Mater. Res. 37 127 (2007). 66 G. Gault, F. Vurpillot, A. Bostel, A. Menand, and B. Deconihout, Appl. Phys. Lett. 86 94101 (2005).

PAGE 189

189 67 E. W. Mller and K. Bah adur, Phys. Rev. 102 624 (1956). 68 E. W. Mller Phys. Rev. 102, 618 (1956). 69 E. W. Mller, J. A. Panitz, and S. B. McLane, Rev. S ci. Instrum. 39 83 (1968). 70 J. A. Panitz, Rev. Sci. Instrum. 44 1034 (1973). 71 A. Cerezo, T. J. Godfrey, and G. D. W. Smith, Rev. Sci. Instrum. 59 862 (1988). 72 T. F. Kellly, P. P. Camus, D. J. Larson, L. M. Holzman, S. S. Bajikar, Ultramicroscopy. 62 29 (1996). 73 T. F. Kelly, D. J. Larson, K. Thompson, R. L. Alvis, J. H. Bunton, J. D. Olsen, and B. P. Gorman, Annu. Rev. Mater. Res. 37 681 (2007). 74 S. Koelling, M. Gilbert, J. Goossens, A. Hikavyy, O. Richard, and W. Vandervorst, Appl. Phys. Lett. 95 144106 (2009). 75 J. H. Bunton, J. D. Olson, D. R. Lenz, and T. E. Kelly, Microsc. Microanal. 13 418 (2007). 76 M. Tsukada, H. Tamura, K. P. McKenna, A. L. Shluger, Y. M. Chen, T. Ohkubo, and K.Hono, Ultramicroscopy 111 567 (2011). 77 Y. M. Chen, T. Ohkubo, and K. Hono, Ultramicroscopy. 111 562 (2011). 78 K. Hono, T. Ohkubo, Y. M. Chen, M. Kodzuka, K. Oh ishi, H. Se pehri Amin, F. Li, T. Kinno, S. Tomiya, and Y. Kanitani, Ultramicroscopy. 111 576 (2011). 79 C. Oberdorfer, P. Stender, C. Reinke and G. Schmitz, Microsc. Microanal. 13 342 (2007). 80 K. Thompson, J. H. Booske, D. J. Larson and T. F. Kelly, Appl. Phys. Lett. 87 052108 (2005). 81 K. Inoue, F. Yano, A. Nishida, T. Tsunomura, T. Toyama, Y. Nagai and M. Hasegawa, Appl. Phys. Lett. 92 103506 (2008). 82 K. Inoue, F. Yano, A. Nishida, H. Takamizawa, T. Tsunomura, Y. Nagai and M. Hasegawa, Appl. Phys. Lett. 95 043502 (2009). 83 K. Thompson, P. L. Flaitz, P. Ronsheim, D. J. Larson and T. F. Kelly, Science 317 1370 (2007).

PAGE 190

190 84 S. Duguay, T. Philippe, F. Cristiano, and D. Blavette, Appl. Phys. Lett. 97 242104 (2010). 85 O. Cojocaru Miredin, D. Mangelinck, K Hoummada, E. Cadel, D. Blavette, B. Deconihout, and C. Perrin Pellegrino Scripta Mater. 57 373 (2007). 86 O. Cojocaru Miredin, E. Cadel, F. Vurpillot, D. Mangelinck, and D. Blavette, Scripta Mater. 60 285 (2009). 87 O. Cojocaru Miredin, E. Cadel, D. B lavette, D. Mangelinck, K. Hoummada, C. Genevois, and B. Deconihout, U ltramicroscopy. 109 797 (2009). 88 K Hoummada, C. Perrin Pellegrino, and D. Mangelinck, J. Appl. Phys. 106 063511 (2009). 89 J. S. Moore, K. S. Jones, H. Kennel, and S. Corcoan, Ultram icroscopy. 108 536 (2008). 90 M. R. Holzworth, N. G. Rudawski, S. J. Pearton, K. S. Jones, L. Lu, T. S. Kang, F. Ren, and J. W. Johnson, Appl. Phys. Lett. 98 122103 (2011). 91 M. Gilbert, W. Vandervorst, S. Koelling, and A. K. Kambham, Ultramicroscopy. 1 11 530 (2011). 92 A. K. Kambham, J. Mody, M. Gilbert, S. Koelling, and W. Vandervorst, Ultramicroscopy. 111 535 (2011). 93 J. D. Brown, R. Borges, E. Piner, A. Vescan, S. Singhal and R. Therrien, Sol. State Electron. 46 1535 (2002). 94 B. Boudjelida, M. C. Simmonds, I. Gee and S. A. Clark, Appl. Surf. Sci. 252 5189 (2006). 95 N. V. Edwards, M. D. Bremser, J. T. W. Weeks, R. S. Kern, R. F. Davis and D. E. Aspnes, Appl. Phys. Lett. 69 2065 (1996). 96 T. Hashizume, S. Ootomo, S. Oyama, M. Konishi and H. H asegawa, J. Vac. Sci. Technol. B 19 1675 (2001). 97 T. Hashizume, S. y. Ootomo, R. Nakasaki, S. Oyama and M. Kihara, Appl. Phys. Lett. 76 2880 (2000). 98 X. L. Wang, D. G. Zhao, J. Chen, X. Y. Li, H. M. Gong and H. Yang, Appl. Surf. Sci. 252 8706 (2006)

PAGE 191

19 1 99 M. R. Johnson, D. A. Cullen, L. Liu, T. S. Kang, F. Ren, C. Y. Chang, S. J. Pearton, S. Jang, J. W. Johnson and D. J. Smith, J. Vac. Sci. Technol. B. 30 062204 (2012). 100 M. Liang and M. E. Law, IEEE T. Comput. Aid. D. 13 1235 (1994). 101 M. E. Law and S. M. Cea, Comp. Mater. Sci. 12 289 (1998). 102 J. Curie and P. Curie, CR. Hebd. Acad. Sci. 91 249 (1880).

PAGE 192

192 BIOGRAPHICAL SKETCH Monta Raymond Holzworth, Jr. was born in Silicon Valley in California which probably contributed to his desire to study s emiconductors. He was educated in Irmo, SC where he graduated from Dutch Fork High School in 2003. He enrolled in the Department of Material s Science and Engineering at the University of Florida in August 200 3 as a National Merit Scholar. In May 2007, h e graduate d Magna cum Laude with a Bachelor of Science in Materials Science and Engineering from the University of Florida Following the completion of his undergraduate education, he enrolled in graduate school at the University of Florida in the Departm ent of Materials Science and Engineering where Kevin S. Jones was his adviser. In May 2009, he earned a Master of Science in Materials Science & Engineer from the University of Florida Finally, he received his Ph.D. from the University of Florida in Aug ust 2013 in Materials Science and Engineering studying the reliability and failure of AlGaN/GaN HEMTs. During his time at UF he watched the teams and two BCS National Championship Football teams. In 20 07, he was th e youngest American finisher of the 1200 km Paris Brest Paris bike ride, and completed the route again in 2011. Most importantly, he married Jennette Van Dien in November 2009. Together they rescued their cat son Tyson in August 2010, and de livered their human son Monta III in November 2012. In his free time he enjoys hanging out sporting, and gaming with his wife and sons.